Content uploaded by Carlos Becker Westphall
Author content
All content in this area was uploaded by Carlos Becker Westphall on Aug 30, 2015
Content may be subject to copyright.
AFIN
2015
The
Seventh
International Conference on Advances
in Future Internet
ISBN:
978
-
1
-
61208
-
428
-
2
August 23
-
28, 2015
Venice
,
Italy
AFIN
201
5
Editors
Eugen Borcoci, University "Politehnica" of Bucharest (UPB), Romania
Carlos Becker Westphall, U
niversity of Santa Catarina, Brazil
1 / 79
AFIN
2015
Foreword
The
Seventh
International Conference on Advances in Future Internet
(
AFIN
2015
)
,
h
eld
be
tween
August 23
-
28, 2015
in
Venice
,
Italy
,
continued a series of events dealing with advances
on
future Internet mechanisms and services.
We are i
n the early
stage of a revolution on what we call Internet now. Most of the
design principles and deployments, as well as originally intended services,
reached
some
technical limits and we can see a tremendous effort to correct this. Routing must be more
i
ntelligent, with quality of service consideration and 'on
-
demand' flavor, while
the access
control schemes should allow multiple technologies yet guarantying the privacy and integrity of
the data. In a heavily distributed network resources, handling asset
and resource for
distributing computing (autonomic, cloud, on
-
demand) and addressing management in the next
IPv6/IPv4 mixed networks require special effort for designers, equipment vendors, developers,
and service providers.
The diversity of the Internet
-
based offered services requires a fair handling of
transactions for financial applications, scalability for smart homes and ehealth/telemedicine,
openness for web
-
based services, and protection of the private life. Different services have
been developed an
d are going to grow based on future Internet mechanisms. Identifying the
key issues and major challenges, as well as the potential solutions and the current results paves
the way for future research.
We take here the opportunity to warmly thank all the me
mbers
of the
AFIN
2015
Technical Program C
ommittee
,
as well as the numerous reviewers. T
he creation of such a
high
quality conference program would not have been possible without their involvement. We also
k
indly thank all the authors who
dedicated much of
their time and efforts to cont
ribute to
AFIN
2015
. We truly believe that, thanks to all these efforts, the final conference program consisted
of top quality contributions.
Also, this event could
not have been a reality without the support of many individ
uals,
organizations, and sponsors. We are gratefu
l to the members
of the
AFIN
2015
organizing
committee for their help in handling the logistics and for their work to make this professional
meeting a success.
We hope that
AFIN
2015
was a successful inter
national forum for the exchange of ideas
and results between academia and industry and for the promotion of
pro
gress
in
the field of
Future I
nternet.
We are convinced that the participants found the event useful and c
ommunications very
open.
We hope
Venic
e
provided a pleasant environment during the conference and everyone
saved some time for exploring this beautiful city.
AFIN
2015
Chairs
:
Jun Bi, Tsinghua University, China
2 / 79
Eugen Borcoci, University Politehnica of Bucharest, Romania
Petre Dini, Concordia
University
-
Montreal, Canada / China Space Agency Center
-
Beijing,China
3 / 79
AFIN
2015
Committee
AFIN Advisory Chairs
Petre Dini, Concordia University
-
Montreal, Canada / China Space Agency Center
-
Beijing, China
Eugen Borcoci, University Politehnica of Bucharest, Romania
Jun Bi, Tsinghua University, China
AFIN 2015 Technical Pro
gram Committee
Rocío Abascal Mena, Universidad Autónoma Metropolitana
-
Cuajimalpa, México
Marie
-
Hélène Abel, University of Technology of Compiègne, France
Alessandro Aldini, University of Urbino "Carlo Bo", Italy
Javier A. Barria, Imperial College London
, UK
Henri Basson, University of Lille North of France (Littoral), France
Khalid Benali, LORIA
-
Université de Lorraine, France
Jun Bi, Tsinghua University, China
Alessandro Bogliolo, University of Urbino, Italy
Eugen Borcoci, University Politehnica of Buc
harest, Romania
Christos Bouras, University of Patras and Research Academic Computer Technology Institute, Greece
Tharrenos Bratitsis, University of Western Macedonia, Greece
Antonio Brogi, University of Pisa, Italy
Chin
-
Chen Chang, Feng Chia University,
Taiwan
Grzegorz Chmaj, University of Nevada
-
Las Vegas, USA
Hauke Coltzau, Fernuniversität in Hagen, Germany
Javier Cubo, University of Malaga, Spain
Maurizio D'Arienzo, Seconda Università di Napoli, Italy
Jiangbo Dang, Siemens Corporation | Corporate Tec
hnology Princeton, USA
Guglielmo De Angelis, CNR
-
IASI, Italy
Sagarmay Deb, Central Queensland University, Australia
Gayo Diallo, University of Bordeaux Segalen, France
Daniel Díaz
-
Sánchez, University Carlos III
-
Madrid, Spain
Sudhir Dixit, HP Labs India
-
Bangalore, India
Jonas Etzold, Fulda University of Applied Sciences, Germany
Florian Fankhauser, TU
-
Wien, Austria
Wu
-
Chang Feng, Portland State University, USA
Alex Galis, University College London, UK
Ivan Ganchev, University of Limerick, Ireland
Rosar
io G. Garroppo, Università di Pisa, Italy
Christos K. Georgiadis, University of Macedonia, Greece
Apostolos Gkamas, Higher Eccelesiastic Academy Vellas of Ioannina, Greece
William I. Grosky, University of Michigan
-
Dearborn, USA
Vic Grout, Glyndwr Universit
y, U.K.
Adam Grzech, Wrocław University of Technology, Poland
Puneet Gupta, Infosys Labs, India
4 / 79
Dongsoo Han, Korea Advanced Institute of Science and Technology(KAIST), Korea
Sung
-
Kook Han, Won Kwang University, Republic of Korea
Ourania Hatzi, Harokopio Un
iversity of Athens, Greece
Hiroaki Higaki, Tokyo Denki University, Japan
Pin
-
Han Ho, University of Waterloo, Canada
Tobias Hoßfeld, University of Würzburg, Germany
Li
-
Ling Hung, Aletheia University, Taiwan
Sandor Imre, Budapest University of Technology and
Economics, Hungary
Norihiro Ishikawa, Komazawa University, Japan
Hanmin Jung, Korea Institute of Science and Technology Information (KISTI), South Korea
Vana Kalogeraki, Athens University of Economics and Business, Greece
Alexey M. Kashevnik, St.Petersbur
g Institute for Informatics and Automation of the Russian Academy of
Sciences (SPIIRAS), Russia
Beob Kyun Kim, Electronics and Telecommunications Research Institute (ETRI), Korea
Changick Kim, Korea Advanced Institute of Science and Technology (KAIST)
-
Da
ejeon, Korea
Mario Koeppen, Kyushu Institute of Technology, Japan
Samad S. Kolahi, Unitec Institute of Technology, New Zealand
Nicos Komninos, Aristotle University of Thessaloniki, Greece
Christian Kop, Alpen
-
Adria
-
Universitaet Klagenfurt, Austria
George
Koutromanos, National and Kapodistrian University of Athens, Greece
Annamaria Kovacs, Goethe University Frankfurt am Main
-
Institute for Computer Science, Germany
Liping Liu, University of Akron, U.S.A.
Jean
-
Pierre Lorré, Linagora, France
Maode Ma, Nanyan
g Technological University, Singapore
Olaf Maennel, Tallinn University of Technology, Estonia
Massimo Marchiori, University of Padua and Atomium Culture, Italy
Brandeis H. Marshall, Purdue University, USA
Francisco Martin, University of Lisbon, Portugal
B
isharat Rasool Memon, University of Southern Denmark, Denmark
Rao Mikkilineni, C3DNA, USA
Debajyoti Mukhopadhyay, Maharashtra Institute of Technology, India
Julius Mueller, Technical University Berlin, Germany
Juan Pedro Muñoz
-
Gea, Polytechnic University o
f Cartagena, Spain
Masayuki Murata, Osaka University, Japan
Prashant R.Nair, Amrita Vishwa Vidyapeetham University, India
Nikolai Nefedov, ETH Zürich, Switzerland
Jose Nino
-
Mora, Carlos III University of Madrid, Spain
António Nogueira, University of Aveiro
, Portugal
Flavio Oquendo, IRISA
-
University of South Brittany, France
Andreas Papasalouros, University of the Aegean, Greece
Alexander Papaspyrou, adesso mobile solutions GmbH, Germany
Giuseppe Patane', CNR
-
IMATI, Italy
Przemyslaw Pochec, University of N
ew Brunswick, Canada
Alain Pegatoquet, University of Nice
-
Sophia Antipolis, France
Graciela Perera, Northeastern Illinois University, USA
Agostino Poggi, Università degli Studi di Parma, Italy
Elaheh Pourabbas, National Research Council of Italy
-
Institut
e of Systems Analysis and Computer
Science "Antonio Ruberti", Italy
5 / 79
Emanuel Puschita, Technical University of Cluj
-
Napoca, Romania
Mayank Raj, Missouri University of Science and Technology, USA
Khairan Dabash Rajab, Najran University, Saudi Arabia
Jaime Ra
mírez, Universidad Politécnica de Madrid, Spain
Torsten Reiners, Curtin University, Australia
Jelena Revzina, Transport and Telecommunication Institute (TTI), Latvia
Simon Pietro Romano, University of Napoli 'Federico II', Italy
Gustavo Rossi, La Plata Nat
ional University, Argentina
Kouichi Sakurai, Kyushu Unversity, Japan
Demetrios G Sampson, University of Piraeus, Greece
Abdolhossein Sarrafzadeh, Unitec Institute of Technology
-
Auckland, New Zealand
Hira Sathu, Unitec Institute of Technology, New Zealand
Hiroyuki Sato, University of Tokyo, Japan
Hans D. Schotten, University of Kaiserslautern, Germany
Bernd Schuller, Jülich Supercomputing Centre, Germany
Omair Shafiq, University of Calgary, Canada
Asadullah Shaikh, Najran University, Saudi Arabia
Dimitrios
Serpanos, University of Patras and ISI, Greece
Nikolay Shilov, St. Petersburg Institute for Informatics and Automation of the Russian Academy of
Sciences (SPIIRAS), Russia
Dorgham Sisalem, Iptelorg/Tekelek, Germany
Michael Sheng, The University of Adelaid
e, Australia
Vasco Soares, Instituto de Telecomunicações / Polytechnic Institute of Castelo Branco, Portugal
José Soler, Technical University of Denmark, Denmark
Kostas Stamos, University of Patras, Greece
Tim Strayer, BBN Technologies, USA
Maciej Szostak
, Wroclaw University of Technology, Poland
Alessandro Testa, National Research Council (CNR) & University of Naples "Federico II", Italy
Brigitte Trousse, INRIA Sophia Antipolis, France
Steve Uhlig, Queen Mary, University of London, UK
Hans van den Berg, T
NO / University of Twente, Netherlands
Rob van der Mei, CWI Centrum Wiskunde en Informatica, The Netherlands
Costas Vassilakis, University of Peloponnese, Greece
Sodel Vázquez Reyes, Universidad Autónoma de Zacatecas, Mexico
Massimo Villari, University of
Messina, Italy
Maurizio Vincini, Universita' di Modena e Reggio Emilia, Italy
Hans Weigand, University of Tilburg, Netherlands
Jozef Wozniak,
Gdańsk University of Technology, Poland
Toshihiro Yamauchi, Okayama University, Japan
Zhixian Yan, Samsung Research, USA
Jianguo Yao, Shanghai Jiao Tong University, China
Chai Kiat Yeo, Nanyang Technological University, Singapore
Martin Zelm, INTEROP
-
VLab
, Belgium
Fan Zhao, Florida Gulf Coast University, U.S.A.
6 / 79
Copyright Information
For your reference, this is the text governing the copyright release for material published by IARIA.
The copyright release is a transfer of publication rights, which allows IARIA and its partners to drive the
dissemination of the p
ublished material. This allows IARIA to give articles increased visibility via
distribution, inclusion in libraries, and arrangements for submission to indexes.
I, the undersigned, declare that the article is original, and that I represent the authors of t
his article in
the copyright release matters. If this work has been done as work
-
for
-
hire, I have obtained all necessary
clearances to execute a copyright release. I hereby irrevocably transfer exclusive copyright for this
material to IARIA. I give IARIA p
ermission or reproduce the work in any media format such as, but not
limited to, print, digital, or electronic. I give IARIA permission to distribute the materials without
restriction to any institutions or individuals. I give IARIA permission to submit th
e work for inclusion in
article repositories as IARIA sees fit.
I, the undersigned, declare that to the best of my knowledge, the article is does not contain libelous or
otherwise unlawful contents or invading the right of privacy or infringing on a propri
etary right.
Following the copyright release, any circulated version of the article must bear the copyright notice and
any header and footer information that IARIA applies to the published article.
IARIA grants royalty
-
free permission to the authors to dis
seminate the work, under the above
provisions, for any academic, commercial, or industrial use. IARIA grants royalty
-
free permission to any
individuals or institutions to make the article available electronically, online, or in print.
IARIA acknowledges th
at rights to any algorithm, process, procedure, apparatus, or articles of
manufacture remain with the authors and their employers.
I, the undersigned, understand that IARIA will not be liable, in contract, tort (including, without
limitation, negligence),
pre
-
contract or other representations (other than fraudulent
misrepresentations) or otherwise in connection w
ith the publication of my work.
Exception to the above is made for work
-
for
-
hire performed while employed by the government. In t
hat
case, copyright to the material remains with the said government. The rightful owners (authors and
government entity) grant unlimited and unrestricted permission to IARIA, IARIA's contractors, and
IARIA's partners to further distribute the work.
7 / 79
Table of Contents
The City Tour Service in Mobile Ad-hoc Group
HyunKyung Yoo, YoungMee Shin, and Changsup Keum
1
An Analysis of the Relationship between Human Personality and Favored Location
Ha Yoon Song and Eun Byul Lee
5
A Web-based System to Manage Primary Students’ Homework -- A Case Study for the Teaching of Portuguese
Miguel Candeias, Maria Gentil Reis, Emanuel Peres, Joaquim Escola, and Manuel J. C. S. Reis
11
Embodying Latent Requirements with Unexperienced Attractions through Selection of Travel Point Photographs
Momoko Kato, Kenjo Yasui, and Hiromitsu Shimakawa
18
Estimating Consumer Inclination for Agricultural Products from Web Browsing History
Riki Tatsuta, Sunao Nakanishi, Yusuke Kajiwara, and Hiromitsu Shimakawa
25
Secure Communication Between OpenFlow Switches and Controllers
Dominik Samociuk
31
Cloud Assisted Live Video Streaming over DHT Overlay Network
Pheng Un Lim and Hwangkyu Choi
37
An Intelligence System based on Social Web Mining and its Application in Health Care in Hong Kong
Kin Keung, Lai, Juan Shi, and Gang Chen
41
Energy- and Priority-Aware Traffic Engineering for Content-Centric Networking
Xu Ling and Yagyu Tomohiko
46
On Security-Effective Mobility-QoS Management Scheme in Heterogeneous Mobile Networks
Hyeungwoo Lee, Jae-Young Choi, and Jongpil Jeong
54
Delay Constrained ARQ Mechanism for MPEG Media Transport Protocol Based Video Streaming over Internet
Hong-rae Lee, Tae-jun Jung, Kwang-deok Seo, and Chang Ki Kim
60
On Cost-Reduced Channel Changing for Mobile IPTV Services in LTE-Advanced Systems
Hyeungwoo Lee, Jae-Young Choi, and Jongpil Jeong
65
Powered by TCPDF (www.tcpdf.org)
1 / 1 8 / 79
The City Tour Service in Mobile Ad-hoc Group
HyunKyung Yoo and Changsup Keum
Communication Service Platform Research Section
Electronics and Telecommunications Research Institute
(ETRI)
Daejeon, Republic of Korea
hkyoo@etri.re.kr, cskeum@etri.re.kr
YoungMee Shin
Energy IT Technology Research Section
Electronics and Telecommunications Research Institute
(ETRI)
Daejeon, Republic of Korea
ymshin@etri.re.kr
Abstract— With the recent explosive growth of mobile devices,
the demand of high quality contents in mobile network is
increasing. To ensure the low delay contents sharing in mobile
network, there is a need to deploy service platform which
provides the intelligent services in the localized ad-hoc group.
The service platform can provide a common structure to create
and support the direct services at user’s edge node. Mobile
subscribers come together in same area, establish a temporal
ad-hoc group and they connect to the same Wi-Fi access point.
People who have common interest can communicate for
sharing contents and information. For them, service platform
can support the storing, caching, and management of contents,
in order to share the high quality contents among proximity
group in mobile network. The feature of proximity group
services has instantly generated social relation. In this paper,
we propose the architecture of service platform for providing a
city tour service. Also, we provide the design and the
development of the proximity services which is composed of a
contents sharing service and a proximity chatting service. Our
platform and service can be applied to the campuses and malls
as well as tour group.
Keywords-service platform; ad-hoc community; proximity
service; city tour service; contents sharing
I. INTRODUCTION
The demand of mobile data services which create, store,
make use of contents is increasing [1]. To deliver the high
quality contents services in mobile network, it is necessary to
deploy caching server in the edge node and provide the
contents sharing service among the localized ad-hoc group. It
is profitable to support the low delay contents through an
edge node and reduce the traffic to core network [2-5].
Two major technologies that allow wireless connection
between mobile devices are Bluetooth and Wi-Fi. Both are
available on most modern mobile phones, tablets, laptops,
and other electronic devices. Wi-Fi can achieve much higher
transmission speed and wider range than Bluetooth. So Wi-
Fi technology has dominated for most high-bandwidth
internet-based applications [6].
For the user-centric service, the service platform is
essential to provide a common structure to create and
provide the direct services at user’s edge node over Wi-Fi [7-
9]. Also, mobile subscribers come together in same area such
as city tour, establish an ad-hoc group, and get a same
interest. For them, the intelligent service platform can
support the storing, caching, and management of contents, in
order to share the high quality contents among proximity
group in mobile network.
In this paper, we propose the architecture of service
platform for providing a city tour service in mobile ad-hoc
group. And, we provide the design and the development of
the proximity services which is composed of a contents
sharing service and a proximity chatting service.
This paper is organized as follows. Section II presents the
architecture of service platform for providing a city tour
proximity service. We explain the functions of social sharing
unit, content sharing unit, chat server, content server, service
discovery unit, and mobile device. And we describe the
service activation sequence flow.
In Section III, we present the city tour proximity service
which is composed of a contents sharing service and a
proximity chatting service. And we show the proximity
chatting service flow. Finally, Section IV summarizes and
concludes the paper.
II. SERVICE PLATFORM FOR MOBILE ADHOC
SERVICE
The city tour services are based on the localized ad-hoc
group connected to Wi-Fi. These groups have instantly
generated social relation, on the basis of interest rather than
intimate relations. The proximity group members can share
the information of same topics at the same place at the same
time. Their shared contents are photo, audio, video, text, and
file. Shared contents must be guaranteed user’s anonymity
and considered a lifetime because of a member’s temporal
relation.
This proximity group service can be very useful in
moving vehicle. For example, the tour guide and tourists can
share the information of tour spot, restaurant, hotel, and
shopping place, by making the dynamic proximity
community.
A. Service Platform Architecture
For the mobile ad-hoc group services, we designed the
service platform which supports the content management and
the social sharing.
1Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
9 / 79
The proposed service platform architecture is shown in
Figure 1. There are two parts of the mobile device and the
service platform. The service platform consists of Social
Sharing Unit, Content Sharing Unit, Chat Server, Content
Server, and Service Discovery Unit. And Mobile Device has
Client App and Discovery Agent.
Figure 1. Service Platform Architecture
Users receive a city tour service in mobile device. The
Discovery Agent of a mobile device discovers the
connection information of the mobile device in the network.
It gets the Access Pointer IP address and the service
connection URL (Uniform Resource Locator). Client App of
mobile device performs the service request and response
according to user interaction.
Social Sharing Unit and Content Sharing Unit in service
platform are web application for user interaction. They
request the chat service or content service, and receive HTTP
(Hypertext Transfer Protocol) based response from Chat
Server or Content Server. They are composed of JavaScript,
HTML5, and CSS (Cascading Style Sheets). JavaScript
handles events, and HTML5 and CSS are the view part of
web applications.
Chat Server provides the chat member management
function based on REST interface and the chat message
management function based on TCP socket connection. The
chat member management function manages the member
nicknames to guarantee an anonymity among the proximity
ad-hoc community. It is the unique member identifier and it
cannot be duplicated in the same Wi-Fi. The chat member
management function performs the management to add and
delete members. The chat message management function
posts and stores the members’ chat messages. If a member
leaves a Wi-Fi network, the member’s temporal relation
disappears.
Content Server provides the contents management to
cache, store, and delete contents, and the contents
consumption like download or streaming. Also it supports
the keyword based search.
Service Discovery Unit manages the service lists
deployed to service platform and returns the service
connection URL to mobile device.
B. Service Activiation Sequence Flow
For using the city tour service, user mobile device must
be connected to Wi-Fi access point, and it gets a service
connection URL. And then, mobile users are offered the city
tour services from a client app deployed to mobile device.
Figure 2 shows these service activation sequence flow
between mobile device and service platform. The related
functions of service activation are service discovery unit,
social chat unit, discovery agent, and client app.
The flows of connection to Wi-Fi network and the
service activation are as follows.
Figure 2. Service Activation Sequence Flow
- Discovery Agent of mobile device is on and waits to
listen to signal.
- Discovery Agent gets the IP address of access point.
- Discovery Agent connects to Service Discovery Unit
of service platform.
- Service Discovery Unit returns the service
connection URL to Discovery Agent.
- Mobile user executes a Client App.
- Discovery Agent transfers the service connection
URL to client app.
- Client App requests service connection URL.
- User confirms service connection and uses the
service.
III. CITY TOUR SERVICE
We have implemented a city tour service which shows
the temporal social relation features of service platform.
Service development device was done on the Samsung
Galaxy Note 8.0, running Android version 4.3 (Jelly Bean).
The demonstrated city tour service is composed of a
contents sharing service and a proximity chatting service.
2Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
10 / 79
In contents sharing service, members can upload, list, and
search the shared contents. The proximity chatting service
provides the social sharing space to share information
messages and contents among members who have same
interest.
Figure 3 shows the proximity chatting service UI of city
tour service. We define the intelligent agent which manages
the chat member presence. When members enter the social
sharing space with nickname, they can see the member lists
and post the information messages. They upload contents
and view shared contents and contents information.
Figure 3. Proximity Chatting Service
The detail flow of proximity chatting service is shown in
Figure 4. It shows the relation among components of service
platform. A member creates the unique nickname and starts
the proximity chatting service. Social Sharing Unit requests
to get chat member lists and content lists. The chat server
returns member lists and the content server returns the
content lists.
When the members input the chat message for
information sharing, the chat server transfers the messages to
all members. While chatting, members can upload and share
the related contents.
The advantage of the city tour service provides the useful
information based on the temporal ad-hoc relation. People
connected to the same Wi-Fi form an ad-hoc community,
and people who have common interest can communicate for
information sharing. Even if they are moving, service
platform maintains the proximity community services.
Also, for privacy protection, when a person who
uploaded content leaves a Wi-Fi network, the shared
contents will be considered expired and disappear. It reflects
the feature of ad-hoc services. Also, the shared contents
among temporal members need to have an anonymity
guaranteed.
Figure 4. Proximity Chatting Service Flow
IV. CONCLUSION
In this paper, we presented the service platform for
providing a city tour service in the mobile ad-hoc group. Our
service platform supports the storing, caching, and
management of contents, in order to share the high quality
contents among the proximity ad-hoc group in mobile
network.
We developed the city tour service, which is composed
of a content sharing service and a proximity chatting service.
The advantage of the city tour service provides the useful
information based on the temporal ad-hoc community.
The lifetime of shared contents is not permanent because
of a member’s temporal relation. The chatting service based
on user nickname can guarantee anonymity among ad-hoc
members.
For further study, we have a plan to extend the service
platform for the proximity intelligent advertisement and
recommendation. It would be taken into account that
platform provides the environment for advertisement and
recommendation lifecycle management.
ACKNOWLEDGMENT
This work was supported by ETRI R&D Program funded
By the Government of Korea (15ZI1410, Development of
Technologies for Proximity, Real-time, and Smart Service
Recommendation Platform).
3Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
11 / 79
REFERENCES
[1] Cisco white paper, “Cisco Visual Networking Index:Global
Mobile Data Traffic Forecast Update, 2014–2019,” tech. rep.,
http://www.cisco.com, accessed Feb. 27th, 2015.
[2] Namiot, Dmitry, and Manfred Sneps-Sneppe. "Proximity as a
service." Future Internet Communications (BCFIC), 2012 2nd
Baltic Congress on. IEEE, 2012.
[3] Heredia, Edwin, et al. "Contextual proactivity for media
sharing scenarios in proximity networks." Consumer
Communications and Networking Conference (CCNC), 2014.
[4] Wang, Yufeng, et al. "Survey on mobile social networking in
proximity (MSNP): approaches, challenges and architecture."
Wireless networks, pp. 1295-1311, 2014.
[5] Salmon, Justin Lewis, and Rong Yang. “A Proximity-Based
Framework for Mobile Services,” IEEE International
Conference on Mobile Services, pp. 124-131, 2014.
[6] Kooh, Thomasgeorgescyrille, Qin Lv, and Shivakant Mishra.
"Attribute Based Content Sharing in Mobile Adhoc Networks
of Smartphones over Wifi." Computer Communications and
Networks (ICCCN), 2012.
[7] Service Delivery Platform (SDP) definition in Wikipedia,
http://en.wikipedia.org/wiki/Service_delivery_platform
[8] Pitkänen, Mikko, et al. "SCAMPI: service platform for social
aware mobile and pervasive computing." ACM SIGCOMM
Computer Communication Review, pp. 503-508, 2012.
[9] Rodriguez-Sanchez, M. C., et al. "GAT: Platform for
automatic context-aware mobile services for m-tourism."
Expert Systems with Applications, pp. 4154-4163, 2013.
4Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
12 / 79
An Analysis of the Relationship
between Human Personality and Favored Location
Ha Yoon Song, Eun Byul Lee
Department of Computer Engineering,
Hongik University, Seoul, Korea
Email: hayoon@hongik.ac.kr, quftmxk@gmail.com
Abstract—It is a long-held belief that human personality affects
the preferred locations of humans. We used statistical regression
analysis in this research to examine the relationship between
human personality and favored locations. The personality data
is represented in accordance with the Big Five Factor (BFF) and
location data was acquired from smartphone apps used by the
volunteers. Instead of using raw location data, a total of 11,154
location datum was collected and then subject to further, more
detailed categorization. The categorized location data were then
classified into a higher-level aggregation. Finally, the relationship
and the statistical results from regression analysis are presented,
alongside categorized and classified location data. A summary of
the relationship is provided, whereas a Coefficient of Determinant
(CoD) is used for the results of the regression analysis.
Keywords–Personality; Favorite location; Regression analysis,
Personality-location relationship, Big Five Factors.
I. INTRODUCTION
The actions of humans can be predicted by time, personal-
ity factors, occupation, gender, and so on. Among these factors,
human personality factors are typically fundamental in the
prediction of human behavior, especially location preference.
In this research, the effect that personality factors exert on
a human’s preferred location was examined. The five-factor
model (FFM), a representative theoretical framework for un-
derstanding human personality, is comprised of the Big Five
Factor (BFF), which can be obtained by using the Big Five In-
ventory (BFI) questionnaire set [1]. For this research, the BFFs
of the 14 volunteer participants were obtained anonymously.
Nowadays, smartphones can be used to easily collect loca-
tion or position data. Numerous smartphone functions includ-
ing Global Positioning System (GPS), GLONASS, and indoor
positioning systems are used for positioning, and humans can
use smartphone apps to easily check-in wherever they are
located. The most popular location apps are Foursquare and
Swarm. This location data can be categorized into higher-level
categories using the categories-id; for example, Vietnamese
restaurant and Japanese restaurant can be categorized in the
higher-level category of Restaurant, while Coffee house and
Milk tea shop can be categorized in Beverage store. These
categories can be identified by the categories-id. However, for
the purpose of detailed research, lower-level categories are also
used in this paper.
Regression analysis [2] was a helpful tool for examining
the relationship between BFF and the categorized locations.
Collected data sets were formed as variables for the re-
gression analysis. In this research, an examination of the
relationship between human personality and location prefer-
ence was conducted. Apart from other research, regression
analysis on personality data and location data were utilized.
Mobile services which are strongly dependent on location can
utilize the benefit of this research. For example, the result
of this research will be applied for recommendation system
which recommends a specific travelers attraction to a person
with certain personality. A possible scenario is that a travel
recommendation system based on our approach, which can
match the traveler’s personality with a property of travelers’
attractions and can recommend a specific travelers attraction
which fits the traveler’s personality.
This paper is organized as follows. Section II will discuss
about related works on location and personality. In Section III,
personality data, location data, and the categories of locations
are described as parts of the preparation for the regression
analysis. Section IV will analyze the results of the regression
analysis and will present examples of the way in which the
BFF affects locations. Section V will present the conclusion
and possible future research topics.
II. RELATED WORKS
There has been little research into how our personality
relates to the categories of place that we visit. There are
mainly two reasons. The first one is related with personal
information protection, meaning peoples do not like to totally
reveal personal information and only volunteers help to collect
personality data without any legal problem. The second one
is that the use of mobile device is not general at all, even
though the smartphones are widely spread. This restricts the
collection of location data. Therefore, only volunteers with
mobile devices can provide their personality data and location
data. Despite of these barriers of data collection, several related
researches have found in last few years.
The relationship between social network and human per-
sonality are conducted in [3]. The effect of human personality
on Facebook usage is researched. Self-report of participants
are used to extract object criteria and measurements from
Facebook data, and the result shows that strong relationship
between human personality and actions on Facebook.
Location Based Social Networks (LBSN) are being started
in this research area nowadays. One of the research [4]
is mostly related to our research. Personality factors and
foursquare check-ins are investigated and utilized, and the re-
lationship is founded. Conscientiousness, Openness, and Neu-
roticism are found related to specific locations in this research.
5Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
13 / 79
As a result of this research, the combination of personality
factors and LBSN can explain the human personality.
Regression analysis on personality data and location data
can be rarely found as shown in [5] in order to predict the
future human location with five volunteers’ data and Back
Propagation Network (BPN). An opposite of this research
can be found in [6] which deduces personality factors form
location data of volunteers. BFFs can be deduced from sets of
location data also by BPN. Apart from the other researches,
our research is a direct match between BFF (personality) and
location categories. Personality data and location data collected
by volunteers themselves are used with minimal treatment of
the raw data.
In the next sections we will present our method to reveal
the relationship between personality and location preference.
III. PERSONALITY AND LOCATION DATA PREPARATION
Two different assortments of location data will be ad-
dressed in this section. The first method is a classification of lo-
cation data. From the location data source, several information
such as frequency of visit, name of location, category-id will be
extracted. The second method is the categorization of location
data. From the location data source, several information such
as frequency of visit, name of location, name of location cat-
egories will be extracted. Note that categorization of location
data is a more ramification of location data comparing to
classification of location data.
Total fourteen volunteers provided their personality and lo-
cation information. All volunteers are anonymous for personal
information protection and called by their numbers. Most of
the volunteers are university students and are in their twenties.
University students tend to go out lively and visits university,
restaurant, and other places frequently. BFF of each volunteer
is identified by BFI. Due to the nature of volunteers that they
are university students, places in university campus are taken
special care and categorized as detail as possible.
TABLE I. BFF OF VOLUNTEERS
O C E A N
Volunteer 1 3.30000 3.88889 3.25000 3.66667 2.62500
Volunteer 2 3.60000 3.33333 2.75000 3.22222 2.75000
Volunteer 3 2.70000 3.22222 3.25000 2.66667 2.75000
Volunteer 4 4.33333 3.12500 2.25000 3.20000 2.88889
Volunteer 5 4.20000 4.33333 3.50000 3.55556 2.62500
Volunteer 6 4.00000 3.66667 4.00000 3.88889 2.75000
Volunteer 7 3.50000 3.77778 3.37500 3.22222 3.00000
Volunteer 8 2.20000 3.44444 3.00000 3.11111 2.62500
Volunteer 9 2.60000 2.77778 3.37500 3.11111 2.62500
Volunteer 10 3.30000 2.88889 3.12500 3.11111 3.25000
Volunteer 11 3.40000 3.22222 3.37500 3.33333 3.12500
Volunteer 12 3.10000 3.66667 3.37500 3.22222 3.50000
Volunteer 13 3.40000 3.55556 3.62500 2.88889 2.50000
Volunteer 14 3.80000 4.00000 3.12500 3.77778 2.25000
A. Personality data
Research has been conducted on the various theories that
seek to represent human personality. Among these theories,
McCrae and Costa’s [7] research on BFF is one of the foremost
studies on personality representation, whereby the five factors
are Openness (O), Conscientiousness (C), Extraversion (E),
Agreeableness (A), and Neuroticism (N). This distinguished
Five Factor Model (FFM) is composed with BFF which can
be obtained by Big Five Inventory (BFI) [8]. In order to obtain
BFF of volunteers, volunteers answer the questionnaire set
called BFI which is usually composed of 44 questions and
the answers are selections of a degree from one to five. The
BFI is designed to calculate every five factors of BFF from
a combination of degrees of the 44 answers. The degree
of each five factor is designed to be represented from one
to five. A major merit of BFF is that the factors can be
quantized and it reflects the concept of normal distribution, i.e.,
each quantitative factor is in a numerical form that stands for
positions of normal distribution. Due to the properties of BFF,
corresponding numerical data can be utilized by any algorithm
and, of course, it can be applied to regression analysis as
shown in this research. It is notable that each factor of BFF is
mutually orthogonal, which stands that factors do not interfere
each other and thus can be used as independent variables in
regression analysis.
TABLE II. SAMPLE LOCATION DATA OF VOLUNTEER 1
CoV Location name categories-id
75 Hongik Univ.T-819 4bf58dd8d48988d19e941735
4 Starbucks 4bf58dd8d48988d1e0931735
3 CGV 4bf58dd8d48988d180941735
2 Hongik Univ. 4bf58dd8d48988d1a8941735
1 Everland 4bf58dd8d48988d182941735
Table I shows BFFs for 14 volunteers, and the values can
be used to easily interpret the personality of a human. For
example, in Table I, Volunteer 4 has the highest Openness
rating among the 14 volunteers, implying that this volunteer
is highly creative in the arts, highly intellectual, curious, and
adventurous. The other BFF ratings of volunteer 4 imply
planning skills, propensity for action, and diligence in regard to
work. Volunteer 6 has the highest Extraversion rating, implying
an energetic, verbose, and positive-minded personality.
The above BFF of Volunteers data set will be used for this
research henceforth.
TABLE III. SAMPLE CLASSIFICATION OF LOCATIONS OF
VOLUNTEER 1
categories-id Name ICC
4bf58dd8d48988d16d941735 Beverage Store 5622
4bf58dd8d48988d110941735 General Restaurant 5611
4bf58dd8d48988d19e941735 Institutions of Education 8530
4bf58dd8d48988d1e2931735 Museum and Historic Site Managing 9022
4bf58dd8d48988d1d0941735 Other Restaurant 5619
4bf58dd8d48988d182941735 Theme Park Operations 9121
B. Classification of location data
Location data, as different from positioning data, represent
conceptual locations such as home, restaurant, school, and
so on. Positioning data is typically numerical data, including
latitudinal and longitudinal pairs. The location data in this
research were collected using the smartphone apps named
Foursquare and Swarm. Volunteers checked-in on their pre-
ferred app whenever they visited a meaningful location. The
app then collated the location data based on the source data
provided by the volunteers, including count of visit (CoV),
the location’s name, and the categories-id. Table II shows a
sample of location data from volunteer 1 as an example of the
data representation of this study.
In Table II, both the formal and numerical forms of the
location information are presented; for example, two visits
6Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
14 / 79
TABLE IV. COUNT OF VISIT (CoV) TO THE LOCATIONS AND ICC OF VOLUNTEER 1
ICC 8530 5611 5619 5291 5622 5621 5914 9900 9022 4712 9112 9121
CoV 30 22 13 6 6 5 3 3 2 1 1 1
TABLE V. VOLUNTEERS’ LOCATION DATA
Volunteer FI RB TI Restaurant Bar BS Bank PAI
1 0.15008 0.04384 0.04384 0.27487 0.02529 0.10118 0.00337 0.00000
2 0.16484 0.06154 0.01319 0.32527 0.02637 0.09890 0.00439 0.00219
3 0.03030 0.09848 0.05303 0.63636 0.02272 0.02272 0.00000 0.00000
4 0.29339 0.01222 0.02119 0.18744 0.00570 0.16299 0.00081 0.00081
5 0.30158 0.00000 0.02116 0.19047 0.01587 0.04232 0.00000 0.00000
6 0.12931 0.02586 0.02874 0.28735 0.02298 0.10344 0.00000 0.00000
7 0.13500 0.00500 0.07000 0.18500 0.01500 0.01000 0.00000 0.00000
8 0.27003 0.04748 0.01187 0.16617 0.01186 0.04154 0.00593 0.00593
9 0.24915 0.20478 0.02048 0.15017 0.00682 0.07508 0.01023 0.00341
10 0.18143 0.02109 0.08439 0.22784 0.03375 0.10126 0.00000 0.00000
11 0.26042 0.01042 0.01042 0.22916 0.03125 0.03819 0.00000 0.00000
12 0.20952 0.11429 0.04286 0.09523 0.00476 0.04285 0.00952 0.00476
13 0.18705 0.03597 0.18705 0.17266 0.01438 0.01438 0.00000 0.02877
14 0.03904 0.14571 0.28619 0.15262 0.04303 0.02951 0.00399 0.00860
Volunteer LS IE Hospital CH LM SF TP BB
1 0.00337 0.30860 0.00337 0.01349 0.01855 0.00505 0.00505 0.00000
2 0.00000 0.27033 0.00879 0.00879 0.01098 0.00219 0.00219 0.00000
3 0.00000 0.09848 0.02272 0.01515 0.00000 0.00000 0.00000 0.00000
4 0.00000 0.28280 0.00244 0.00733 0.00570 0.00733 0.00489 0.00489
5 0.00000 0.39153 0.00000 0.01587 0.01587 0.00529 0.00000 0.00000
6 0.00287 0.37931 0.00862 0.00000 0.00862 0.00287 0.00000 0.00000
7 0.00000 0.57000 0.01000 0.00000 0.00000 0.00000 0.00000 0.00000
8 0.00000 0.41246 0.00296 0.00890 0.00890 0.00000 0.00593 0.00000
9 0.00000 0.19453 0.01706 0.00682 0.01706 0.03413 0.00000 0.01023
10 0.00000 0.23628 0.00843 0.01687 0.02109 0.06751 0.00000 0.00000
11 0.00000 0.29513 0.01736 0.05556 0.00347 0.03819 0.01041 0.00000
12 0.00000 0.42381 0.00476 0.02857 0.01428 0.00000 0.00476 0.00000
13 0.00000 0.28777 0.00719 0.01438 0.03597 0.00000 0.00719 0.00719
14 0.00292 0.07254 0.06947 0.04980 0.04857 0.02797 0.01644 0.00353
were made to a location named Hongik Univ.T-819, which
has a predefined categories-id of 4bf58dd8d48988dla8941735.
According to the Standard Industrial Classification, higher
categories can be created by merging subcategories using
the categories-id; for example, Vietnamese restaurant and
Japanese restaurant can be merged into the higher category
of Restaurant. In another case, Coffee house and Milk tea
shop can be merged into Beverage Store, which is a higher
level category also classified in the Standard Industrial Clas-
sification. Using Standard Industrial Classification prepared
by the National Tax Service and the Industry Classification
Code (ICC), locations can be classified into higher-level ICC
categories. Table III shows several samples of the location
classifications using the data from Volunteer 1. Table III
shows the categories-id, the names of the classified cate-
gories, and the ICC; for example, Hongik Univ. T-819’s
categories-id is 4bf58dd8d48988d19e941735, and after it was
merged into Institutions of Education, a higher classification,
the corresponding ICC became 8530. Similarly, categories-
id 4bf58dd8d48988d182941735 was merged into Theme Park
Operations with an ICC of 9121.
Table IV shows extracts of CoV values to each ICC; for
example, one visit was made to the ICC location of 4712 and
22 visits were made to the ICC location of 5611. Table II,
Table III, and Table IV show the sequence of the location-
classification process for Volunteer 1.
It was necessary to arrange all of the data from the
14 volunteers in the same way. Table V summarizes the
classification results of every volunteer. The meanings of the
location-classification acronyms in Table V are:
• FI = Foreign Institutions
• RB = Retail Business
• TI = Travel Industry
• BS = Beverage Store
• PAI = Public Administration Institutions
• LS = Leasing Service
• IE = Institutions of Education
• CH = Concert Hall
• LM = Library, Museum
• SF = Sports Facilities
• TP = Theme Park
• BB = Beauty Business
Every number in Table V represents the ratio of visits
according to ( 1), as follows:
V isitingRatio(loc) =
count of visit to(loc)
total count of visit
(1)
where loc is one of the locations in Table V.
For example, the CoV of Volunteer 1 to Foreign Institute is
89, while the total CoV is 593; therefore, visiting ratio (FI) =
0.15008. As shown in Table V, Volunteer 1 has a visiting ratio
of 0.15008 for Foreign Institute. In the case of Concert Hall,
Table V shows zero values for Volunteer 6 and Volunteer 7,
while other volunteers have a non-zero visiting ratio, meaning
that more than one visit was made.
7Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
15 / 79
C. Categorization of location data
From the source check-in data, each location category
was parsed, extracting the CoV, the location name, and the
categories-name. Table VI shows several of the extracted
samples, including one example where 58 visits were made
to a location named Hongik Univ.T, which is categorized as
University Buildings.
Instead of considering each location with the standards
of location classification, the CoV was aggregated for each
categories-name as shown in Table VII. Table VII shows CoV
according to non-overlapped categories-name. Also, Table VII
has more detailed information for Volunteer 1 than Table III;
for example, bus terminals are clearly separated from train
stations, i.e., categorized rather than classified. An interpre-
tation of Table VII shows that Volunteer 1 visited Home 80
times, University Buildings 76 times, and Korean Restaurant
39 times. The sample data shown in Table VI and Table VII
was collected only from Volunteer 1. Table VII shows the
location data only from volunteer 1, meaning that there is a
total of 14 similar tables for each volunteer.
In sum, 96 categories-names were derived from the data of
the 14 volunteers with some overlapping of categories-names.
After the Standard Industrial Classification and ICC, location
categories with a total CoV less than 10 for all volunteers were
merged into Etc. category. The final data set consists of 55
categories-names of the 14 volunteers and provided dependent
variables for the regression analysis.
TABLE VI. SAMPLE OF CATEGORIZED DATA OF VOLUNTEER 1
CoV Location name Categories-name
58 Hongik Univ.T University Buildings
5 CGV Ori Theater
4 Songtan yeongbillu Chinese Restaurant
3 Shinsegae Department Store Department Store
3 TongtongPig BBQ shop
TABLE VII. VOLUNTEER 1: REPRESENTATIVE CATEGORIES-NAME
AND CORRESPONDING COUNT OF VISIT
Categories-name CoV Categories-name CoV
University Lab 83 Language School 4
Home 80 Train Station 4
University Buildings 76 Art Museum 4
Cafe 60 Bank 4
Korean Restaurant 39 Fast Food Shop 4
etc. 39 Bus Stop 3
University Library 21 Cosmetics Shop 3
Dessert Shop 17 Sushi Shop 3
Outdoor Activity 15 Subway 3
Bar 12 Grocery Store 3
Chinese Restaurant 11 Shopping Centre 3
Snack Shop 11 Clothing Store 2
Noodles Restaurant 8 American Restaurant 2
Fried Chicken Shops 8 Hospital 2
BBQ Shop 8 Convenience Store 2
Japan Restaurant 8 Museum 1
Italy Restaurant 6 Gymnasium 1
Bakery 6 Salad Section 1
Theater 6 Electronics Stores 1
Airport 5 University Stores 1
Rest Area 5 Asian Restaurant 1
Department Store 5 Pizzeria 1
Bus Terminal 5 Seafood Shoppe 1
IV. RELATIONSHIP RESULT ANALYSIS
In this section, two sorts of location data set from Sec-
tion III will be regarded as two sets of dependent variables.
Corresponding results will be addressed.
A. Regression analysis on location classifications
A basic regression analysis was conducted to further de-
velop the examination. Each volunteers’ BFF values, as shown
in Table I, was regarded as an independent variable. The
location information, as shown in Table V, was regarded as a
dependent variable. The regression analysis showed the effect
of personality on location preference. The basic result is shown
in Table VIII using Coefficient of Determination (CoD).
Table VIII contains the following acronyms:
• CoD = Coefficient of Determination
• IE = Institutions of Education
• PAI = Public Administration Institutions
The relationship between each variable can be represented
by CoD. In Table IX, the range of CoD and the interpretations
of relationships are presented. In cases where the CoD value
is higher than 0.4, the relationship was considered to be high.
According to this criteria, a case with a CoD value higher than
0.4 was regarded as one in which the personality and location
are meaningfully related.
TABLE VIII. RESULTS OF REGRESSION ANALYSIS FOR CLASSIFIED
LOCATIONS
Location Classification CoD Location Classification CoD
Leasing Service 0.71391 Library, Museum 0.31678
IE 0.64162 Retail Business 0.26152
Restaurant 0.55895 Travel Industry 0.25720
Bank 0.52616 Foreign Institutions 0.23086
Beauty Business 0.49252 Theme Park 0.22845
PAI 0.45437 Bar 0.17817
Sports Facilities 0.39169 Beverage Store 0.15551
Hospital 0.32293 Art Viewing Facilities 0.09206
TABLE IX. RELATIONSHIP DETERMINED BY COEFFICIENT OF
DETERMINATION
Range of CoD Interpretation
0.9-1.0 Very High
0.7-0.9 High
0.4-0.7 Relatively High
0.2-0.4 Low
0.0-0.2 Rarely Related
Table VIII shows location classifications and the cor-
responding CoD. The location classifications of Restaurant,
Bank, Public Administration Institutions, Leasing Service, In-
stitutions of Education, and Beauty Business show CoD values
higher than 0.4, meaning that only 6 classifications showed a
meaningful relationship between personality and location.
It may be meaningful to see the effect of the subcategories
of the location classifications. The next subsection will show
the results for location categories.
B. Regression analysis on location categories
The regression analysis on location categories found a
higher number of meaningful relationships. BFF values, as
shown in Table I, were also used as an independent variable,
while the total 55 location categories were used as a dependent
variable. The regression analysis showed that 31 location
categories had a CoD higher than 0.4, indicating a meaningful
relationship between personality and location.
8Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
16 / 79
Table X shows a selection of meaningful results from the
regression analysis and, in addition to the CoD, a coefficient
for Openness, Conscientiousness, Extraversion, Agreeableness,
and Neuroticism (O, C, E, A, N) was used.
The coefficient of O, C, E, A, N shows the degree of
personality’s effect on corresponding location categories. First,
the absolute value of each coefficient stands for the strength
of affection: positive coefficient means positive effect, while
negative coefficient means negative effect.
The criteria of each coefficient are determined to be relative
when the absolute value of the coefficient is more than 0.01,
and strongly related when the absolute value coefficient is
more than 0.1. Table X contains a column titled RELATION
which represents the effect of personality on the location
categories. The RELATION column shows the final results of
this research. As a factor of meaningfulness, the plus sign (+)
stands for the positive relation to the corresponding location,
and the minus sign (-) stands for the negative relation to the
corresponding location. For the strong positive relation, (++)
is used, while the strong negative relation is indicated with the
use of (- -). In addition, p-value must be considered. Once p-
value is greater than 0.05 for a relation, the personality factor
will be statistically disregarded and will not be presented in
RELATION column.
As an example, University Buildings have an O, C, E,
A, N coefficient of -0.13274, 0.07956, 0.01043, 0.00477,
0.21439, respectively. Openness and Neuroticism, which are
also two factors of BFF, show an absolute value more than
0.1; therefore, - -O and ++N show that Openness strongly
affects the location of University Buildings positively and
Neuroticism strongly affects University Buildings negatively.
Conscientiousness and Extraversion have coefficient values
more than 0.01 but lower than 0.1; therefore, +C and +E are
also applicable to the location category of University Building.
Overall, the corresponding relationship column may contain -
-O, +C, +E, and ++N. However, +C and +E must be excluded
since their corresponding p-values are greater than 0.05.
Some of the RELATION columns are remained blank
because the corresponding relation is negligible, whereby the
coefficients of O, C, E, A, N are less than 0.01. The corre-
sponding locations may be related to BFF since CoD is bigger
than 0.4; however, the corresponding coefficient for each BFF
is less than 0.01. For example, in the case of Museum, the
coefficients for O, C, E, A, N are 0.00250 ， 0.0008 ， 0.00402,
-0.00828, -0.00541, respectively. Even though Agreeableness
has the largest absolute coefficient value, implying a negative
relationship, it is not shown.
In terms of restaurants, categories such as Korean Restau-
rant, Asian Restaurant, and Japanese Restaurant sometimes
showed a strong relationship with BFF, while American
Restaurant and Italian Restaurant showed a more minor rela-
tionship with BFF. This phenomenon might have been affected
by CoV, which suggests that more data needs to be collected.
As shown in Table VII, the highest CoVs can be seen for
Home and University, which is unsurprising considering that
most of the volunteers were students at the time of the study.
However, in Table X, University-related locations are present,
while a Home location is not. The CoD for Home is less than
0.4, so maybe the location Home has a neutral relationship with
human personality. Such a large CoV for University, University
Buildings, University Library, University Lab, and University
Stores should be categorized and more closely investigated.
Table X has the highest CoD for University Buildings of
0.87732, and O and N are RELATED for University Buildings.
A similar BFF pattern could possibly be found for workers’
offices. In the case of Fast Food Shop, which has a CoD of
0.75860, negative Extraversion have relations but they are not
strong.
From the results in Table X, Openness, Extraversion and
Neuroticism are affective factors for location visit while Con-
scientiousness is the negligible factor for location visit. Maybe
the more data will reveal more relationship between BFF and
locations since it cannot be guaranteed that our data covers
all possible locations or covers all possible combinations of
personality factors.
V. CONCLUSION
Regression analysis is usually an acceptable tool to use
when deducing statistical fitness in a linear relationship be-
tween independent and dependent variables. BFF values were
used as independent variables in our research, while catego-
rized and classified location data were regarded as dependent
variables. The result of the regression analysis is meaning-
ful and, for several locations, the effect of personality was
revealed. It is meaningful that we used a scientific method,
instead of relying on conjecture, to identify the effect of
personality on several specific locations. We also identified
an aspect of the functional relationship between human per-
sonality and location preference.
Our research can be applied to many of related areas
in order to increase imposed value of each area, such as
enhancement of mobility model [9][10], personal travel rec-
ommendation system [11], mobile computing [12] and so on.
The key is personalized future Internet, especially for location
based social network related area.
An implication of this relationship emerged after the basic
stage of our research, when we had obtained 2,842 location
datum. Eventually, the total of 11,154 location datum was used
to provide more meaningful research results. It is possible
that, if more volunteer participants check-in at more locations,
further insight into the personality-location relationship will be
gained.
However, our research is likely to have biased results in two
reasons. Even though we collected enough number of locations
visited, only 14 volunteers are participating and we have a
small set of BFFs comparing to the size of location data. It is
an unavoidable bias and will be solved with more volunteers
are involved as we continue our research. Sometimes the bias
can be intentional supposing the travel recommendation system
which we mentioned as an application, where travelers can be
considered mostly. The second bias is that most of our volun-
teers are university students, which leads to lots of check-ins
to university buildings. Due to this bias, however, we obtained
meaningful results on the relationship between personality and
university buildings. Since we continue collecting more BFFs,
we expect that more distinct relationship can be found.
The more proper method of our research may be stepwise
regression. Independent application of independent variable
from BFF by stepwise regression will give more clear result
9Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
17 / 79
TABLE X. THE EFFECT OF PERSONALITY TO LOCATIONS WITH COEFFICOENT OF DETERMINATION
Location CoV CoD O C E A N RELATION
University Buildings 741 0.87732 -0.13274 0.07956 0.01043 0.00477 0.21439 - -O, ++N
Fast Food Shop 55 0.75860 0.00467 -0.01153 -0.02299 -0.00340 -0.00628 -E
Korean Restaurant 690 0.72413 0.08739 -0.03577 -0.03540 -0.02518 -0.02823 +O
University Library 104 0.66507 -0.02990 0.02783 0.00124 -0.02349 0.01764 -O
Asian Restaurant 50 0.64682 0.01263 -0.00389 -0.02293 -0.00772 0.00011 -E
Seafood Shoppe 31 0.64301 0.00038 0.00223 -0.00106 0.00057 -0.00158
Sushi Shop 31 0.64136 0.00074 -0.00010 -0.00411 0.00259 -0.00146
Cosmetics Shop 14 0.63816 -0.00118 -0.00579 0.00513 -0.00517 -0.00851
Japan Restaurant 126 0.60027 0.00224 -0.01117 -0.01006 0.00342 0.00876
Pizzeria 10 0.59626 -2.00E-05 -0.00668 -0.00069 0.00297 -0.00254
Hotel 86 0.57299 0.00773 -0.00104 0.00437 -0.01647 -0.01546 -N
Language School 34 0.57297 -0.01789 -0.01246 -0.00626 0.01823 -0.00212 -O
Bank 42 0.56085 -0.00595 0.00112 -0.00181 0.00366 0.00048
BBQ Shop 209 0.54467 0.02409 0.00055 0.00260 -0.02265 -0.00241 +O
Snack Shop 79 0.53491 -0.00561 -0.00752 -0.00728 0.01758 0.00432 +A
University Lab 731 0.52897 0.12886 0.08422 0.10599 -0.07828 -0.00644
Electronics Stores 18 0.52038 0.00352 0.00499 -0.00526 -0.00484 0.00448
Beauty Salon 24 0.51725 0.00083 -0.00273 0.00226 -0.00261 -0.00528
Department Store 35 0.51175 -0.00407 -0.00902 -0.00134 0.00555 -0.00569
Airport 103 0.51101 0.00615 0.00204 0.00314 -0.01302 -0.01522 -N
Rest Area 81 0.50978 0.00911 -0.00300 0.01236 -0.01191 -0.01270
Outlet Store 11 0.48241 -0.00200 -0.00883 0.00328 0.00314 -0.00608
Chicken Shops 90 0.48083 0.00429 -0.00713 -0.00655 0.00452 0.00644
Public Institution 53 0.47681 -0.00358 0.00170 -0.00223 0.00301 -0.00098
Museum 16 0.47463 0.00250 0.00081 0.00402 -0.00828 -0.00541
University Stores 14 0.45565 -0.00135 -0.00779 0.00133 0.00766 -0.00356
Pharmacy 53 0.44116 -9.90E-05 -0.00812 -0.00120 0.00504 -0.00706
American Restaurant 48 0.43308 0.00320 -0.00392 -0.00216 0.00355 0.00161
Italy Restaurant 61 0.43121 -0.00162 -0.00147 0.00241 0.00909 0.00281
Salad Section 14 0.42998 -0.00087 0.00070 -0.00059 0.00158 0.00343
Convenience Store 279 0.42796 -0.00051 0.00194 -0.01514 0.01295 -0.00944
about the effect of each personality factors on location prefer-
ence based on the orthogonality of each five factor.
One more possible research would include temporal in-
formation as well as location information. The check-in time
and check-in duration will provide more information about the
mobility pattern of each volunteer and our understanding of
the personality-location relationship will become more precise.
The inclusion of other personal factors in future research,
including occupation, income, and gender, studies will give
keener insight into this relationship.
ACKNOWLEDGMENT
This work was supported by the National Research Foun-
dation of Korea(NRF) grant funded by the Korea govern-
ment(MEST) (NRF-2012R1A2A2A03046473).
REFERENCES
[1] O. P. John and S. Srivastava, “The big five trait taxonomy: History,
measurement, and theoretical perspectives,” Handbook of personality:
Theory and research, vol. 2, no. 1999, 1999, pp. 102–138.
[2] H. Smith, N. R. Draper, and H. Smith, Applied regression analysis.
Chapman and Hall., 1981.
[3] Y. Amichai-Hamburger and G. Vinitzky, “Social network use and
personality,” Computers in human behavior, vol. 26, no. 6, 2010, pp.
1289–1295.
[4] M. J. Chorley, R. M. Whitaker, and S. M. Allen, “Personality
and location-based social networks,” Computers in Human Behavior,
vol. 46, 2015, pp. 45–56.
[5] S. Y. Kim and H. Y. Song, “Predicting human location based on human
personality,” in Internet of Things, Smart Spaces, and Next Generation
Networks and Systems. Springer, 2014, pp. 70–81.
[6] ——, “Determination coefficient analysis between personality and
location using regression,” in International Conference on Sciences,
Engineering and Technology Innovations, 2015, pp. 265–274.
[7] P. T. Costa and R. R. McCrae, “Four ways five factors are basic,”
Personality and individual differences, vol. 13, no. 6, 1992, pp. 653–
665.
[8] L. R. Goldberg, “The structure of phenotypic personality traits.” Amer-
ican psychologist, vol. 48, no. 1, 1993, p. 26.
[9] D. Alberg, M. Last, and S. Elnekave, “A spatio-temporal simulation
model for movement data generation,” in Data Mining Workshops,
2008. ICDMW’08. IEEE International Conference on. IEEE, 2008,
pp. 320–325.
[10] D. Guo and W. Cui, “Mining moving object trajectories in location-
based services for spatio-temporal database update,” in Sixth Interna-
tional Conference on Advanced Optical Materials and Devices. Inter-
national Society for Optics and Photonics, 2008, pp. 71 432M–71 432M.
[11] Y. Doytsher, B. Galon, and Y. Kanza, “Storing routes in socio-spatial
networks and supporting social-based route recommendation,” in Pro-
ceedings of the 3rd ACM SIGSPATIAL International Workshop on
Location-Based Social Networks. ACM, 2011, pp. 49–56.
[12] W.-C. Peng and M.-S. Chen, “Developing data allocation schemes by
incremental mining of user moving patterns in a mobile computing
system,” Knowledge and Data Engineering, IEEE Transactions on,
vol. 15, no. 1, 2003, pp. 70–85.
10Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
18 / 79
A Web-based System to Manage Primary Students’ Homework
A Case Study for the Teaching of Portuguese
M. Candeias
1
, M.G.A.D. Reis
2
, E. Peres
3
, J.J. Escola
1
, M.J.C.S. Reis
4
1
University of Tr´as-os-Montes e Alto Douro (UTAD), Vila Real, Portugal.
E-mails: {jmpcandeias, mgentil.reis} @gmail.com, {eperes, jescola, mcabral}@utad.pt
2
Escolas Jo˜ao de Ara´ujo Correia, Centro Escolar da Alameda, Peso da R´egua
3
INESC-TEC and UTAD.
4
IEETA and UTAD.
Abstract—Technology has profoundly changed the way we learn
and live. Internet provides a new paradigm, setting forth that
education and communication approach is truly more complex
than ever before. Here, we present a web-based system to support
students’ homework. In order to prove its usefulness, we present
the results achieved with the application of a set of exercises to
help improving Portuguese (mother tongue) language skills of
4th-grade primary pupils. This set of digital exercises proved to
be the child’s favorite, rather than their counterpart exercises
in paper form, also showing a fair more positive attitude from
the students’ point of view. A simple survey also showed that
students do prefer the use of the system here presented, when
compared to other more traditional ways of practicing. We also
believe that the prompt feedback about the exercises correctness,
together with the training provided by the different exercises
sets about the same subject, besides the exploitation of video,
color, sound, etc., positively reinforce the diverse child’s senses,
definitely contributing to capture and motivate the child.
Keywords–Web-based homework system; improving mother
tongue teaching/learning; primary education
I. INTRODUCTION
Technology has profoundly changed the way we learn and
live. This relationship appears to be quite complex, in that
IT, and especially socially- and technologically-rich learning
environments, seems to both require and foster skills and
learning. Internet provides a new paradigm, setting forth that
education and communication approach is truly more complex
than ever before.
It is on the basis of such paradigm that we present a case
study where a set of exercises were used in order to improve
the Portuguese (mother tongue) language skills of 4th-grade
primary pupils. Being part of a web-based system to support
students’ homework [1], a set of digital exercises proved to
be the child’s favorite, rather than their counterpart exercises
in paper form, also showing a fair more positive attitude. To
the best of our knowledge, there is no system like the one
presented here available.
The paper is organized as follows. The next section is
used to present some related work. Section III is used to
present some of the advantages of using the web-based system
presented here to help with the student homework, and related
to the Portuguese mother tong teaching/learing. In Section IV,
we present the theoretical model used to support our research.
In Section V, we present an overview of the web-based system,
and also its developing context. In Section VI, we present some
of the main characteristics of the set of templates available
in the system and used to create the exercises offered to the
students during this study, and also the data and some of our
main achievements. Finally, in Section VII, we present the
major conclusions achieved with this work.
II. RELATED WORK
As far as our knowledge, there are no systems using the
same approach and ideas as those used in the system presented
and discussed in this paper. However, there are many web-
sites and tools available for helping the teaching of virtually
everything.
Concerning the class of available web-sites we want to
highlight the following ones. The “Coursera” site/system [2]
offers free courses online from world-class universities. The
topics are varied and lectures are formatted into series of
15-minute-long clips, and students can watch videos at their
convenience and in their own time. “Learnist” is the “Pinterest”
of learning [3]. The interactive platform allows users collect
teaching materials and educational content that are grouped
into “boards”. The platform is equipped with the “Learn it!”
bookmarklet, which enables picking images anywhere on the
web and automatically shoots them to your “Learnist” board.
The “Open Culture” [4] system is a high-quality cultural and
educational media. A collection of 1,100 courses, online, for
free, from Ivy League universities, such as Stanford, UCLA,
Columbia and Oxford University is available. Courses run the
gamut from science and art to math and economics. “Brazilian”
Portuguese is one of the 48 languages available to learn for
free. However, the type of contents available is completely
different from the ones proposed here. “Udacity” [5] is free
education web-site for brainy types, founded by four Stanford
roboticists, and currently offers 11 courses, all of which are
in science and math-related topics. According to the web-
site, plans are underway to expand the curriculum. Udacity
is free of deadlines, free of prerequisites, free of quizzes and
other annoying school stuff. Courses are also free of charge.
TED (Technology, Entertainment, Design), the powerhouse
of jaw-dropping lectures needs no introduction. Now, they’re
bringing their talent into education with an offshoot, Ted-
Ed [6]. A treasure trove of beautifully animated and gripping
videos on a wide array of subjects, such as “The Power
of Simple Words” and “How Many Universes are there”, is
available. All videos are under 10-minutes-long. Supplemental
materials, such as quizzes and discussion questions, are also
available. Unfortunately, we were unable to find any lessons
on Portuguese in the 108,138 lessons available. Effortless
learning based on three pillars science, fun and community
is the basis for the “MemRise” site [7]. Based on scien-
tific methods for implanting new information, “MemRise”
11Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
19 / 79
proves that what may seem impossible is in fact doable. The
“Guardian Wallcharts” are mapped “MemRise” style, teaching
you all about cheeses, herbs and other wonders of life. The
“MentorMob” site/system [8] is like the “YouTube” of learning
materials: users create learning “playlists” from first-rate web-
sites. MentorMob is a community, whose members share and
rank each others’ learning playlists.
There are many “educational web-sites” for kids/students
available on the Internet. In the site “Wonderopolis” [9], a
kid/student may discover a new wonder each day. Wonders are
amazing facts and intriguing questions such as “Why are they
called Lava Lamps?” or “What Badger has a sweet tooth?” can
be answered. Surfing the web-site “Whyville” kids/students
can learn through playing games and socializing [10]. J.K.
Rowling created the site “Pottermore” so kids could read the
books and do interactive features and games [11]. The site
“National Geographic Little Kids” [12] features games, crafts
and recipes, science, videos, and animal information. It is
perfect for 5 years old and under kids. Math and reading
video-like games like math baseball and Mad Libs Junior are
available at the “Funbrain” web-site [13]. Kids/students can
get into cooking on the “Spatulatta” friendly cooking web-
site with lots of videos and recipes [14]. The “NGA Kids”
web-site [15] gives users art adventures and activities from
the United Sates National Gallery of Art. The “Yahoo Kids”
web-site is the least educational of all the choices listed above,
since the site includes both games and videos of all sorts. But,
kids/students can find lots to learn on this interactive web-site
like homework help, learning about science, and access to an
encyclopedia.
Many more educational sites are available on the Internet;
there are even sites with sorted lists of educational sites, such
as those lists presented in [16]. However, all these sites offer
exercises or sets of exercises, which can (and should) be solved
by kids/students, but the teacher cannot change (customize)
them to create her/his own exercises based on the existing
ones, and then offer them to the kids/students. The web-based
system presented here allows this.
The Learning Managements Systems (LMS), such as Moo-
dle, Blackboard, Sakai, Desire2Learn, etc., possess their own
tools for the creation and publication of exercises, and obvi-
ously should be included in the list of web-based systems and
tools for helping building exercises. Also, there are other tools
available on the Internet:
• “Hot Potatoes” [17] — This program is free to ed-
ucators, although registration is required. There are
six types of quizzes; JBC: multiple-choice exercises,
JCloze: gap-fill exercises, JMatch: matching/ordering
exercises, JQuiz:short-answer questions, JMix: jum-
bled sentences. After the creation of the quizzes, Hot
Potatoes will convert the files to HTML, which the
author must then (manually) upload to the server.
• “Quia! Create your own learning activities!” [18] —
There are eight types of games and two types of
quizzes to create at the Quia! site. The user must
be registered before using Quia!. At the Quia! web-
site the user enters the information and keeps track of
the resulting URL. The user can create a quiz from
any computer with Internet access, produce a page,
and then access it again if the user remembers the
URL. All pages stay on the Quia! server. The user
can choose a discipline on the main page to see what
others have done with Quia!
• “QuizStar” [19] — The Quiz Star is an application
that allows bringing together web pages for a series of
lessons. The QuizStar quizzes remain on the TrackStar
server. Registration is needed.
• “Charity Kahn’s JavaScript Quiz Creator” [20] — This
tool creates the code for a simple multiple choice quiz.
The user will need to copy the code and upload it to
the server.
• “interactivetest.com” [21] — Here we can make online
quizzes (up to 100 total questions) for free. For more
than 100 and there is a charge. The quizzes stay on
the interactivetest.com server.
• “Puzzlemaker” [22] — Is a puzzle generation tool for
teachers, students and parents. The user can create
and print customized word search, criss-cross, math
puzzles, and more-using his/her own word lists. The
user must choose a type of puzzle from the list of
ten templates available and make her/his own puzzle
online for classroom or home use.
• “Rubric Generator” [23] — Rubrics are intended to
make grading quicker, clearer, and more objective.
The site contains tools that guide the user through the
process of creating the assessment tools for evaluating
student performance. There are over five hundred
printable rubrics on the web site.
• “Quiz your Friends” [24] — A very simple site that
will guide the user through the creation of a quiz in
four steps. In the final step the user chooses if the
quiz will be sent via Email, Instagram, Facebook or
Twitter, to the group of peoples/friends.
• A list of many more quiz builders can be found in [25].
The big differences, and main characteristics, of the system
presented here and that makes it unique when compared to the
systems/sites listed above are the following:
• the production/creation of exercises is based on exist-
ing templates;
• exercises can only be created based on the existing
templates;
• anyone, whose access was granted by the administra-
tor of the system, may create exercises;
• the publication of every exercise just created is done
immediately and automatically after its creation, and it
will be available for everyone registered in the system;
• the creation/production of a new exercise is based in
three simple steps:
1) choose a template from the list presented in
the form of images containing all the exer-
cises;
2) change the customizable fields in the tem-
plate;
3) choose a name and save the new exercise in
the exercises database;
• there is no need for any programming skills or expe-
rience.
12Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
20 / 79
As can be seen from the list above, the template dictates
the “behavior” of the exercise. The exercise in “it-self” will
run on the computer of the user, i.e., the exercise will not
run in the “server-side” but in the “client-side”. As will be
discussed in Section V (“Overview of the System”), the system
was implemented using HTML 5, JavaScript, Adobe Flash, and
ActionScript. The system will be used to create the exercises
and store them in its internal databases. After that, the system
will work as a typical web-server.
In fact, in seeking to meet the latest requirements and
concepts used by leading standards in distance learning, we are
making efforts to make it possible to download the exercises in
a standard (SCORM and Common Cartridge) in order to make
possible the execution of exercises off-line and subsequent
communication of the results achieved by students at the time
of reconnection to the Internet. However, at the moment, this
is partially implemented.
III. ADVANTAGES OF USING THE WEB-BASED SYSTEM
In the view of several authors it seems that we are
moving to a “digital learning” (e.g., [26]–[28]). According
to Weller [26], there are five factors that favor de use of
the Internet in education: its social acceptance, it facilitates
a sense of control and therefore ownership for educators far
more than previous technologies, the web browser has become
a generic interface, it is both an interactive (do something
with the information rather than just being a passive recipient)
and personal (all learners are not the same) medium, is a
sustaining (e.g., a retailer that uses it to supplement its physical
shops, for instance by home delivery from supermarkets) and
disruptive (e.g., it does not intend web based shopping to
displace its standard retail outlets, but rather it hopes the two
will complement each other, and thus altering the organization
in which it is implemented) technology.
Internet can be used to supplement face-to-face teaching;
for example, many campus based universities use the Internet
not to replace their traditional face-to-face mode of delivery,
but rather as a means of supplementing it (e.g., web pages can
be used to provide additional information, or e-mail be used
as a means of contacting tutors of large courses). Of course
that there are topics that better suit its use (an “Introduction
to Information Technology” course, and a course for teachers
wanting to learn about computer conferencing, are examples
of courses were using the technology effectively forms an
integral part of the academic content of the course) without
forget pedagogic suitability (for example, the resource based
learning, where students can be presented with a wide range of
resources, often external to the university, were students derive
their own learning experience from these resources within
the overall framework of the course). Internet’s usage also
leads to institutional benefit because institutions can benefit
from the use of Internet in education, both on campus and in
reaching new audiences, and from the fact that it can be seen
as a Computer Assisted Learning delivery mechanism—many
courses are using Internet to deliver simulations, visualization
aids and interactive tools.
In order to become more accurate in their work, both
in reading problems and in working out solutions, primary
students need more practice to reinforce what they are learn-
ing, a process which traditionally takes place on paper, and
traditionally are distributed in textbooks or tutoring books.
Unfortunately, these materials are usually designed for aver-
age learners, and it is often difficult to find the best-fitting
content for students with differing abilities and skills. For
example, in Portugal all the classes have only one textbook
designed for all students, but literate students may need a
higher-level tutoring, and illiterate may need a lower-level
tutoring. Therefore, the depth and flexibility of ability gained
from these textbooks are restricted. Some programs based
on e-learning technology may provide personalized contents
for learners by collecting the learning process. But primary
students may become restless and unfocused when staying in
front of computers during long periods of time. As putted by
Warschauer, [27], “New technologies do not replace the need
for strong human mentorship, but, indeed, amplify the role
of such mentorship”. Obviously students must become into
contact with the new teaching/studding tools progressively, in
order for they to become a part of the learning environment
as smoothly as possible.
Teachers, but also parents, have an important word to say
in this process; as noted by Warschauer [27], “the teacher
must be centrally involved, actively instructing and mentoring
students, especially at the initial stages of work on a project.
Unfocused instruction can leave students rudderless, and this
is particularly harmful to at-risk students, such as those with
learning disabilities, limited literacy, and language skills, or
insufficient background knowledge”.
In general, after students have completed practice exercises
or homework, very often they have to wait for their teacher to
check it and provide feedback. These interruptions can reduce
students’ interest in learning as well as learning efficiency.
We believe that the prompt feedback about the exercises
correctness, together with the training provided by the different
exercises sets about the same subject, besides the exploitation
of video, color, sound, etc., positively reinforce the diverse
child’s senses, definitely contributing to capture and motivate
the child.
From the teacher point of view, one advantage of the
system here proposed is time saving. Most of the times, the
teacher writes in the exercise-book of each pupil (at least in the
“blackboard”) the set of homework exercises that s/he should
try to solve. This is a handy work, and consequently very
time consuming. When students return with their homework,
the teacher must read and correct, one-by-one, every exercise-
book, in order to have and give feedback to the pupils. During
this process, the teacher also analyzes each student’s work,
in order to evaluate the student’s progress and potentially
introduce some programme and/or pedagogical adjustments.
Obviously, this analysis should also be made to the global
class.
With the web-based system presented here the teacher
can select different sets of exercises for each student or,
alternatively, propose the same set of exercises to the whole
class, and this simply by creating e-mail lists and maintaining
these lists as needed. The system will then provide the teacher
with statistical data concerning each student and the entire
class. We believe that these data will greatly help the teacher
in evaluating his/her work, and help with the mentioned
adjustments. The selection of different exercises sets may not
directly lead to a time reduction of the teacher’s work, but
its automatic check and the statistical data do. We believe that
13Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
21 / 79
teachers can use this extra time to better understand where and
why some children are having troubles.
IV. PEDAGOGICAL MODEL
The teaching and researching communities agree that it is
through the use of exercises and games that learning becomes
effective, regardless of the model or philosophy used. For
example, proponents of sociocultural theory claim that learning
is primarily a social process mediated through interactions us-
ing tools [29]–[31]. According to Vygotsky, mediation occurs
through the use of ‘semiotic’ and ‘material’ tools. The semiotic
tools include symbols, signs, and spoken languages. Material
tools include such items as pens, spoons, and particularly
networked (Internet) computers. These tools do not simply
facilitate the set of activity that might take place, but they
fundamentally shape and define the type of activities that might
be developed [30]. Seymour Papert proposed the use of tools,
particularly the computer “a mighty education tool”, in helping
in the process of construction of knowledge (given rise to
the “constructionist” theory), adapting the beginnings of the
cognitive constructivism of Jean Piaget in order to a better use
of technology, [32].
Traditionally, the dominant form of learning in schools has
been teacher-directed learning or guided learning: “a trainer
or teacher takes all the relevant decisions and the learner can
and should follow him or her. He decides about the goals of
learning, the learning strategies, the way to measure outcomes
and he takes care of feedback, judgments, and rewards” [33].
Besides guided learning, Simons, Linden and Duffy, [33],
distinguish two other ways of learning: experiential and action
learning. To facilitate and support learners through the gradual
and progressive acquisition of adaptive competence, novel
classroom practices and cultures are needed. These practices
and cultures should create the conditions for a substantial shift
from guided learning toward experiential as well as action
learning, resulting in a balanced and integrated use of the three
ways of learning: constructive, self-regulated, and contextual
or situated [34].
The constructivist view of learning has become com-
mon ground among educational psychologists (see, for ex-
ample, [33], [35]). Constructivism implies that constructive
learning is self-regulated. According to Zimmerman [36], self-
regulation “refers to the degree that individuals are metacog-
nitively, motivationally, and behaviorally active participants in
their own learning process”. According to De Corte [34], “con-
structive and self-regulated learning processes should prefer-
ably happen and be studied in context”, and because learning
is collaborative, the learning efforts are distributed over the
individual student, its partners in the learning environment, and
the (technological) resources and tools that are available [37].
As stated by De Corte [38], “starting as much as possible
from tasks and problems that are meaningful and challenging
for students, learning environments should initiate socially sup-
ported constructive learning processes that enhance students’
cognitive and volitional self-regulatory skills”. It is expected
that students will be able to use their acquired knowledge
and skills to solve situations and problems in everyday life,
something that Bransford and Schwartz [39] call preparation
for future learning.
This is the theoretical model supporting the web-based
system presented here and this research.
V. OVERVIEW OF THE SYSTEM
Before start describing the system a word about its devel-
oping context is in order.
In 2006, the “ESchool” Programme (e.escola— [40]) en-
abled the purchase of laptops by students, form the 5th to
10th grade, and teachers for a price of 150 euro. Later,
the Portuguese government has developed a set of protocols
with Intel, the leading telecommunications operators (Optimus,
TMN, Vodafone, and Zon), Microsoft, “Magic Box” (Caixa
M´agica), and the local members, allowing children of the
primary schools access to portable computers “Magellan” (Ma-
galh˜aes), for a maximum cost of 50 euro [41]. Additionally, the
classrooms have been equipped with interactive “blackboards”.
The existence of computers in the schools is of course
necessary to integrate IT effectively in the learning process,
but it is by no means sufficient. Setting-up infrastructures is
a relatively simple process in comparison with their actual
usage, since both the necessary training and the change of
habits that lead to routine usage are the result of slow and
gradual processes; the need to train, motivate and involve the
teachers, pupils and the community in general towards a cor-
rect use of IT tools entails a number of technical and cultural
difficulties [42], [43]. Additionally, several researchers claim
that it is what teachers think and believe and students’ attitudes
that ultimately shapes the activities in the classroom [44]–[46].
As usually happens, at least in Portugal, the lack of tools
supporting the authoring and automatic checking of exercises
for specific topics in primary education (e.g., check true or
false sentences) drastically reduces the advantages in the use
of e-learning/web-based environments on a larger scale. Some
technical details of the web-based system presented here were
presented elsewhere [47].
The web-based system presented here aims to contribute to
the resolution of problems in two distinct classes: firstly, for the
spreading, construction and sharing of educational resources,
particularly those closely related to the teaching of Portuguese
as mother tongue; secondly, to develop applications more “user
friendly” in terms of the average user, trying to eliminate the
necessary notions of programming that are currently required
by the existing applications.
It is 100% visual (graphical) for the preparation of the
exercises and games, which in turn are developed based on
existing templates.
Obviously the need for using standards when designing
and implementing tools to help in the creation of learning
objects became evident. Only then the exercises and games
(i.e., learning objects) will be truly available for reuse and
sharing, no matter the platform the students/learners are using.
In our view, SCORM (Sharable Content Object Reference
Model) is the most complete standard to be used in the case
presented here (Primary Students’ Homework), because in
its specification it includes standards developed by various
third parties, integrating various standard specifications that
simultaneously enable better interoperability between systems
and sharing of content, and it is widely accepted that it is the
best standard to use with Computer Based Training [48].
The tool has two distinct modules: one related to the
development and personalization of the templates; and an-
other dedicated to the development and personalization of
14Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
22 / 79
the exercises and games. The main blocks of the applica-
tion include: user’s management (registration and authentica-
tion); templates’ management (creation, modification, deletion,
etc.); exercise’s management (creation, modification, etc.);
databases’ management; exercises’ and templates’ download-
ing; configuring/customization; help; and exiting application.
The databases block is able to manage distinct databases,
including the templates’ and exercises’ databases, as well as all
the information related to the registered users. Also, a special
database containing all the images that can be used to produce
templates and exercises is implemented; obviously, the user
may add new images to the database. So, in its current version,
the tool implements the following data tables: users, templates,
exercises, images, classes (used to classify images).
The templates management block of the application is not
fully operational in its current version. In its current version,
only the exercises module is fully implemented. The templates
module is operational but it cannot be used to create or change
an existing template (i.e., it shows the existing templates to
the user and let he/she choose one from the presented list).
We have created only five templates, using other tools, for
demonstration proposes. When a user/teacher wants to create
a new exercise he/she must follow three main steps. First,
choose a template or an existing exercise, from a list presented
in the form of images containing all the exercises. Second,
change the customizable fields in the template. Third, choose
a name and save the new exercise in the exercises database;
this will generate a XML file with information corresponding
to the configuration options of the exercise (exercise wording,
images, correct options/answers, etc.) that will be saved along
with the rest of the exercise. The number of interactions during
the second step will depend on the template. Typically, it will
be necessary to customize the exercise wording (i.e., the text
explaining what is expected for the student to do), the images
that will be used/presented to the student, as well as the correct
answers/options.
In the example presented in Figure 1, besides the fields
listed above, it is also possible to change the text of the
“Verify” and “Try again” buttons and the text displayed when
the answer is correct and when the answer is wrong. Also, it is
possible to choose the degree of help/difficulty the student will
have during the execution of the exercise (“easy”, “medium” or
“hard”). When in the “ease” mode every time a student chooses
an option or enters his/her answer, he/she will have immediate
feedback about its correctness. When in the “medium” mode,
only when he/she presses the “verify” button will get the
feedback about the correct and wrong answers. In the “hard”
mode only when he/she presses the “verify” button will get the
feedback about the exercise correctness, but the student will
have no feedback about the correct and incorrect answers (only
that there are errors, or not). All these interactions are based on
choosing elements form dropdown lists, check boxes, drag and
drop objects/images, among other. In the example presented in
Figure 2, the student is asked to fill in the missing words (“the
gaps”) in order to form correct sentences; this Figure shows
the exercise seen by the student.
The system was implemented using HTML 5, JavaScript,
Adobe Flash, and ActionScript.
We want that in the near future this system may be used in
English-speaking countries, since the only thing that needs to
be changed is its interface. From the exercises point of view,
Figure 1. Creating a sample exercise a sample exercise—fill in the missing
words (“the gaps”) in order to form correct sentences.
Figure 2. Solving a sample exercise a sample exercise—fill in the missing
words (“the gaps”) in order to form correct sentences.
they do not need adjustments, because it all depends on the
templates (the templates should be designed in English, as well
as everything they contain, such as video, images, text, speech,
music, etc.).
VI. SOME RESULTS
Currently, there are five templates that can be used in
the system to create new exercises: three relate to the teach-
ing/learning of Portuguese mother tong, and the other two
relate to the teaching of other concepts (notions of in and
out, animals and fruits, count, etc.). Here we will present and
discuss the results of first three templates.
The results presented in this section report to a set of
exercises used to demonstrate the usefulness of the proposed
web-based system. We follow the methodology proposed by
Yin [49], where the researcher can conduct the research in its
working context.
Bearing in mind the teaching and learning of Portuguese
mother tongue, and the fact that the exercises should serve
as homework for primary students, we have developed three
specific templates: “fill in the gapes”, “true or false”, and
“multiple choice”. In the “fill in the gapes” template, the
pupils are asked to fill the blank spaces in a sentence, or
set of sentences, in order to form correct sentences. This
template includes the possibility to present to the student the
missing words, but not necessarily obeying the order in which
they should be introduced in the sentence(s), i.e., the words
are randomly presented, functioning as extra help. Also, the
teacher can choose between one and five “missing words”; in
15Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
23 / 79
fact, the “gap” can be a complete sentence. In the “true or
false” template, the students are asked to classify a sentence
as being true or false. In the “multiple choice” template, the
students are asked to select the best possible answer out of the
choices from a list.
As presented above in Section V, the degree of
help/difficulty the student will have during the execution of any
of the exercises may be set to three different levels: “easy”,
“medium” or “hard”. For our set, we have choose easy and
medium levels.
We have also developed the corresponding counterpart
paper format of the set of digital exercises. The exercises were
presented to 4th-year (4th-grade) primary students during the
months of January, February and March 2015, in the context of
the project “Little Box of surprises” (Caixinha das surpresas).
Basically there are sets of “little boxes”, each with a set
of ten to twelve copies of the same story, and once/twice a
month a story is chosen (by the teacher in conjunction with
the librarian), and the students go to the library and read
the story. At the end they are asked to respond to a set of
questions, traditionally in paper form, of the types listed above
(filling gapes, true/false, writing sentences/ordering words, and
multiple choice). The sample exercise in Figures 1 and 2 was
produced based on the fairy tale “The Princess and the pea”
by Hans Christian Anderson. The full set is available in [1]. A
“scanned” form of the corresponding counter-part paper format
offered to the students is also available [50].
All the students have answered all the questions both in
paper format and directly in the web-based system using the
computers available at the library. Table I summarizes the
results achieved by the 23 4th-grade students. The students
have answered 12 questions of the type “True & False” (T&F),
one question with 5 “gapes” in the template “fill in the gapes”,
and 6 questions of the type “multiple choice”. This table
presents the number of correct answers. During the application
process of the web-based exercises we gave the students the
possibility to answer correctly all the questions and registered
the number of tries. However, due to space limitations and
because in the paper form students have only one chance to
answer (correctly or not) all the questions, here we present
the results of the first answers (i.e., here, we consider that an
answer was correct only if it was correctly answered during
the first attempt). As can be seen, for the “T&F” exercises the
average value is the same both for paper and web-based types.
For the “gapes” and “choice” types the results were slightly
better for the exercises in paper format. Note that the mode is
the same for the “gapes” exercises, is better for the “choice”
exercises in paper format, and is better for the “gapes” when
using the web-based system.
We also presented the students a very simple survey with
the following questions:
1) The web-based system was easy to use.
2) It was easy to understand.
3) I would have liked more in class use.
4) It was helpful because I had prompt feedback.
5) I would have preferred more homework questions
using it (in place of other assignments).
6) I would have liked more contents of this type on it.
7) It was easier to answer the questions in it.
8) I felt better responding using the computer.
0.5
0.6
0.7
0.8
0.9
1
Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8
Figure 3. Survey results: average answer to each question.
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20
Figure 4. Survey results: average answers from each student.
Figure 3 shows the mean responses to each question of this
survey. The Portuguese version of the survey (actually applied)
can be downloaded [50]. As can be seen from the plots in
this Figure, questions #1, 3 and 8 drew the highest responses,
indicating that students found the system easy to use and feel
comfortable using it. However, question #7 had the lowest
response, showing that they don’t feel that it was easier to
answer the questions using the computer, or at least they don’t
feel that the questions were not of lower difficulty degree. This
seems to be reinforced by the responses to questions #4 and
6. Average answers for each survey participant, presented in
Figure 4, show that some responses seemed to be either too
low or too high, i.e., showing dissatisfaction/satisfaction with
the class/subject itself and not specifically with the web-based
system.
Students’ comments were overall positive. The less positive
comments point to a need to have more homework examples,
and more time in order to gain experience in using the system
and computer. Next are listed some of the comments. “I think
we should use more often the computer”. “I really enjoyed
doing this on the computer where everything was cool and
made me learn a lot.” “I think it was easy to use the computer
because almost hit all the questions at first!”. “I learned to
write on the computer.” “I really enjoyed! It was great fun!”.
“The computer is cool!”. “He liked to use the computer more
often”. “It was fun and also spectacular”. “I wish I had more
time to do so”. “I liked using the computer because it was fun
to learn on the computer”.
16Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
24 / 79
TABLE I. EXERCISES’ RESULTS SUMMARY FOR THE 23 STUDENTS (T&F—TRUE & FALSE; GAPES—FILL IN THE GAPES; CHOICE—MULTIPLE CHOICE;
AVG—AVERAGE; STD—STANDARD DEVIATION; MOD—MODE).
Student
Avg Std Mod
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Paper
T&F 10 8 10 10 8 12 11 11 11 8 9 10 10 11 10 10 10 9 7 10 11 11 11 9.9 1.2 10
Gapes 5 5 3 5 0 5 5 5 4 5 5 5 5 5 5 5 5 5 4 5 5 5 5 4.6 1.1 5
Choice 6 6 6 6 5 6 6 6 6 6 6 6 6 6 5 6 6 3 6 6 5 5 6 5.7 0.7 6
Internet
T&F 11 10 8 9 9 9 11 8 8 8 11 11 10 10 11 12 10 10 11 9 11 11 9 9.9 1.2 11
Gapes 4 5 5 4 3 5 4 4 3 5 3 4 5 5 5 5 5 5 3 5 5 5 5 4.4 0.8 5
Choice 4 4 4 3 4 5 4 5 4 5 5 4 4 5 5 6 4 2 3 5 6 6 6 4.5 1.0 4
In conclusion, we could see that there are no big differences
in the results achieved by the students. However, during the
answers to the exercises students showed a more positive
attitude, happiness and greater openness to the use of the web-
based system.
VII. CONCLUSION AND FUTURE WORK
We have presented a web-based system [1] to support
students’ homework. In order to prove its usefulness, we have
built a set of exercises to help improving Portuguese (mother
tongue) language skills of 4th-grade primary pupils. This set
of digital exercises proved to be the child’s favorite, rather
than their counterpart exercises in paper form, also showing
a fair more positive attitude from the students’ point of view.
A simple survey also showed that students do prefer the use
of the system here presented, when compared to other more
traditional ways of practicing.
We believe that the prompt feedback about the exercises
correctness, together with the training provided by the different
exercises sets about the same subject, besides the exploitation
of video, color, sound, etc., positively reinforce the diverse
child’s senses, definitely contributing to capture and motivate
the child.
Students with disabilities are known for possessing a set
of unique characteristics that hinder their integration in school
and consequently their learning. We believe that students with
disabilities can be the first benefitting from using a system like
the one proposed here.
As stated by Warschauer, [27], “New technologies do not
replace the need for strong human mentorship, but, indeed,
amplify the role of such mentorship”. Obviously that the stu-
dents must become into contact with the new teaching/studding
tools progressively, in order for they to become a part of the
learning environment as smoothly as possible.
From the examples above, we could easily concluded that
the presented templates may be used to teach/learn other
subjects. For example, based on the “true or false” template a
teacher can build a set true or false questions for the teaching
of chemistry. Obviously, the “multiple choice” template may
be used to build exercises for the teaching of any subject.
REFERENCES
[1] http://www.mcabral.utad.pt/apps/. Accessed: 10-08-2015.
[2] https://www.coursera.org/. Accessed: 10-08-2015.
[3] http://learni.st/. Accessed: 10-08-2015.
[4] http://www.openculture.com/. Accessed: 10-08-2015.
[5] https://www.udacity.com/. Accessed: 11-08-2015.
[6] http://ed.ted.com/. Accessed: 11-08-2015.
[7] http://www.memrise.com/. Accessed: 11-08-2015.
[8] http://www.mentormob.com/. Accessed: 11-08-2015.
[9] http://wonderopolis.org/. Accessed: 12-08-2015.
[10] http://www.whyville.net/smmk/nice. Accessed: 12-08-2015.
[11] https://www.pottermore.com/. Accessed: 12-08-2015.
[12] http://kids.nationalgeographic.com/. Accessed: 12-08-2015.
[13] http://www.funbrain.com/. Accessed: 12-08-2015.
[14] http://www.spatulatta.com/. Accessed: 12-08-2015.
[15] http://www.nga.gov/content/ngaweb/education/kids.html. Accessed: 12-
08-2015.
[16] http://www.parenting.com/. Accessed: 10-08-2015.
[17] http://web.uvic.ca/hrd/halfbaked/. Accessed: 13-08-2015.
[18] http://www.quia.com. Accessed: 13-08-2015.
[19] http://quizstar.4teachers.org/. Accessed: 13-08-2015.
[20] http://builder.i.com.com/. Accessed: 13-08-2015.
[21] http://www.interactivetest.com. Accessed: 13-08-2015.
[22] http://www.discoveryeducation.com/free-puzzlemaker/. Accessed: 13-
08-2015.
[23] http://www.teach-nology.com/web tools/rubrics/. Accessed: 13-08-
2015.
[24] http://www02.quizyourfriends.com/. Accessed: 13-08-2015.
[25] http://eleaston.com/quizzes.html. Accessed: 13-08-2015.
[26] M. Weller, Delivering Learning on the Net: the why, what & how of
online education. London and New York: RoutledgeFalmer, Taylor &
Francis Group, 2002.
[27] M. Warschauer, “The paradoxical future of digital learning,” Learning
Inquiry, vol. 1, 2007, pp. 41–49.
[28] K. Moeller, U. Fischer, H.-C. Nuerk, and U. Cress, “Computers in
mathematics education - Training the mental number line,” Computers
in Human Behavior, vol. 48, JUL 2015, pp. 597–607.
[29] L. S. Vygotsky, Mind in society: Development of Higher Psychological
Processes, M. Cole, V. John-Steiner, S. Scribner, and E. Souberman,
Eds. Harvard University Press, 1978.
[30] J. V. Wertsch, “The voice of rationality in a sociocultural approach
to mind,” in Vygostky and education: Instructional implications and
applications of sociohistorical psychology, L. C. Moll, Ed. New York,
NY: Cambridge University Press, 1992, pp. 111–126.
[31] C. A. Bisol, C. B. Valentini, and K. C. Rech Braun, “Teacher education
for inclusion: Can a virtual learning object help?” Computers &
Education, vol. 85, JUL 2015, pp. 203–210.
[32] S. M. Papert, Mindstorms: Children, Computers, and Powerful Ideas.
New York: Basic Books, 1980.
[33] R. J. Simons, J. van der Linden, and T. Duffy, “New learning: Three
ways to learn in a new balance,” in New learning, R. J. Simons,
J. van der Linden, and T. Duffy, Eds. Dordrecht, The Netherlands:
Kluwer Academic Publishers, 2000, pp. 1–20.
[34] E. D. Corte, “Learning from instruction: the case of mathematics,”
Learning Inquiry, vol. 1, 2007, pp. 19–30.
[35] H. Passos, “Flexible contents: Communities of practice and distributed
cognitivism impact in instructional design,” in 3rd International Con-
ference of Education, Research and Innovation (ICERI2010), Chova,
LG and Belenguer, DM and Torres, IC, Ed. Lauri Volpi 6, Valenica,
Burjassot 46100, Spain: IATED-INT Assoc Technology Education A&
Development, 2010, Proceedings Paper, pp. 329–333, 3rd International
17Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
25 / 79
Conference of Education, Research and Innovation (ICERI), Madrid,
Spain, Nov 15-17, 2010.
[36] B. J. Zimmerman, “Dimensions of academic self-regulation: A con-
ceptual framework for education,” in Self-regulation of learning and
performance: Issues and educational applications, D. H. Schunk and
B. J. Zimmerman, Eds. Hillsdale, NJ: Lawrence Erlbaum Associates,
1994, pp. 3–21.
[37] G. Salomon, Ed., Distributed cognition. Psychological and educational
considerations. Cambridge, UK: Cambridge University Press, 1993.
[38] E. D. Corte, “Mainstreams and perspectives in research on (mathemat-
ics) learning from instruction,” Applied Psychology: An International
Review, vol. 53, 2004, pp. 279–310.
[39] J. D. Bransford and D. L. Schwartz, “Rethinking transfer: A simple
proposal with multiple implications,” in Review of research in educa-
tion, A. Iran-Nejad and P. Pearson, Eds. Washington, DC: American
Educational Research Association, 1999, vol. 24, pp. 61–100.
[40] http://www.eescola.net/indexA.aspx. Accessed: 10-08-2015.
[41] http://eescola.pt/e-escolinha/Oferta.aspx. Accessed: 10-08-2015.
[42] M. J. C. S. Reis, G. M. M. C. Santos, and P. J. S. G. Ferreira, “Promoting
the educative use of the internet in the Portuguese primary schools: a
case study,” Aslib Proceedings, vol. 60, no. 2, 2008, pp. 111–129.
[43] R. Shillair, S. R. Cotten, H.-Y. S. Tsai, S. Alhabash, R. LaRose, and
N. J. Rifon, “Online safety begins with you and me: Convincing Internet
users to protect, themselves,” Computers in Human Behavior, vol. 48,
JUL 2015, pp. 199–207.
[44] S. A. Yoon, J. Koehler-Yom, E. Anderson, J. Lin, and E. Klopfer,
“Using an adaptive expertise lens to understand the quality of teachers’
classroom implementation of computer-supported complex systems
curricula in high school science,” Research in Science & Technological
Education, vol. 33, no. 2, May 4 2015, pp. 237–251.
[45] S. Kerckaert, R. Vanderlinde, and J. van Braak, “The role of ICT in
early childhood education: Scale development and research on ICT use
and influencing factors,” European Early Childhood Education Research
Journal, vol. 23, no. 2, Mar 15 2015, pp. 183–199.
[46] A. Taneja, V. Fiore, and B. Fischer, “Cyber-slacking in the classroom:
Potential for digital distraction in the new age,” Computers & Education,
vol. 82, Mar 2015, pp. 141–151.
[47] S. Neto, A. Capela, R. Morais, E. Peres, and M. Reis, “A HTML
5-based tool to help teachers in the creation of exercises and games
for the teaching of mathematical concepts in the primary schools,” in
2nd International Conference on Virtual and Networked Organizations
- Emergent Technologies and Tools (ViNOrg 13), P´ovoa de Varzim,
Portugal, 2013, pp. 1–6.
[48] V. Gonzalez-Barbone and L. Anido-Rifon, “From SCORM to Common
Cartridge: A step forward,” Computers & Education, vol. 54, no. 1,
2010, pp. 88–102.
[49] R. K. Yin, Case study research: Design and methods, 2nd ed. Newbury
Park, CA: Sage, 1984.
[50] http://www.mcabral.utad.pt/HansChristianAnderson/
SapatinhosVermelhos. Accessed: 10-08-2015.
18Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
26 / 79
Embodying Latent Requirements with Unexperienced Attractions
through Selection of Travel Point Photographs
Momoko Kato
Graduate School of Information Science and Engineering
Ritsumeikan University
Shiga, Japan
Email: momochii@de.is.ritsumei.ac.jp
Kenjo Yasui, Hiromitsu Shimakawa
Ritsumeikan University
Shiga, Japan
Email: {kenjo, simakawa}@de.is.ritsumei.ac.jp
Abstract—People who want to travel search for travel points
satisfying their travel requirements. They search for travel points
such that they feel attractions to fulfill the requirements. They
want to search for them easily and quickly. In recent years, there
are many people who use the Internet to collect information on
travel points. They use the Internet because they can collect the
information on the travel point anytime and anywhere. On the
other hand, there are various kinds of information on a lot of
travel points on the Internet. Users have to embody conditions
to specify the travel point they want to visit. True pleasures
of travels are to experience what they have never experienced.
Requirements for travel points to seem to be attractive are quite
unclear. For the recommendation of travel points, people want
a tool to embody easily and quickly the conditions specifying
the travel point they want to visit. Current search engines for
travel points on the Internet utilize the user based collaborative
filtering from travel histories of travelers. However, they cannot
grasp what latent requirements travelers have and what kinds of
attraction they expect, because they analyze travel histories. The
paper proposes a recommendation method which enables users to
embody each potential requirement for their travel beforehand.
The method utilizes photographs of travel points to infer what
latent travel requirement users have and what kinds of attractions
they expect to fulfill the latent requirements. An experiment has
proved that almost two thirds of subjects are satisfied with travel
points recommended using this method. Furthermore, the method
allows all of the subjects to decide on travel points they want to
visit in a shorter time than the method they usually use.
Keywords–Travel; Trip; searching keyword; recommendation;
preference;
I. INTRODUCTION
Some people go on a trip to relax their body and mind
worn out in real life, while others aim to change their moods.
They decide on vacation places based on events which incline
them to trips. After they decide on vacation places, they
would search for details of what they can experience there.
For example, they would check foods specific to the places,
famous sightseeing areas, accommodation facilities, and so on.
The information collection continues until they are satisfied as
long as their time allows. In these years, many people gather
the vacation information with Internet, because they can search
for the travel information anytime and anywhere with it. On
the other hand, there is too much travel information on the
Internet. A user must select appropriate travel information to
decide on a vacation place. It is required for all users to find out
the travel information useful to decide on vacation places, in a
short time and with little effort. Specifying keywords relating
to vacation places to search engines on the Internet, people
planning trips browse the vacation information to decide on
a vacation place with Internet. The information collection
does not need much effort, if users of search engines have
embodied keywords which enable them to narrow candidate
vacation places meeting with their requirements. However, the
information collection is not easy in many cases, because users
are likely to have only unclear requirements in the beginning
of planning of their trips. To make the best use of the tools to
collect travel information, we should support them to embody
keywords representing their vacation requirements in their
minds.
The pleasure of planning trips is the process of chooseing
vacation places, imaging themselves enjoying the places they
search with the Internet. Some users find new vacation goals
or their latent preference for trips, when they attain unknown
vacation information which excites them in the planning. The
new goals or latent preference for trips make it more difficult
to embody keywords. Users with little knowledge on vacation
cannot precisely embody keywords to represent vacation places
meeting their goals and preference. They have little chance
to find suitable vacation places matching their images. On
the other hand, users with a lot of vacation experiences have
often fixed their preference. They are likely to specify same
keywords in all cases. They are disappointed with stereotyped
trip images brought by search results. They fail to expand
their expectation for a trip. Several travel recommendation
systems have been developed with the user-based collaborative
filtering from the travel histories [8] [9] [10]. But, it is the real
pleasure of trips to visit places which provide entertainments
users have never experienced. Users often have new goals
and latent preference for trips. For new goals, the users
have no knowledge to achieve them. For latent preference,
the users often find them in places they visit by chance.
Any existing system using travel histories cannot recommend
vacation places matching with new goals and latent preference,
becuase they are not attained from places they have visited in
their past trips. It is likely to recommend users vacation places
similar to what the users have been to. Users cannot be excited
when the system recommends those vacation places.
This paper proposes a method to embody user requirements
for a trip, which are ambiguous in the beginning of trip
19Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
27 / 79
planning. The method estimates goals and preference of user
for trips, using photographs showing what users can experience
in specific vacation places. In the process of searching for
vacation places using photographs, it is easy for users to
image themselves there. Suppose a user is pleased with a
photograph of a specific vacation place, after imaging herself
there. It is assumed that the photograph implies her goals
and preference for trips, including not only apparent ones
but also the latent ones. The proposed method has collected
various goals and preference from many people, beforehand.
Each photograph matches with some of them. The method
associates terms representing them with the photograph in
advance. It regards associated terms as keywords to embody
her requirements for trips. The method lets the user specify the
keywords in the search bar of existing Internet search engines
for vacation places. It presents vacation places recommended
in high ranking, which enable the user to decide on the
vacation places matching with her goal and preference of trip.
The remaining of the paper is organized as follows. Section
2 describes the present state and problems of the searching
for vacation places with the Internet. The method and the
usability of the photograph are presented in Section 3. Section
4 illustrates experiments to verify the usefulness of the method.
The paper discusses the experiment result in Section 5. Section
6 gives the conclusions and future works.
II. HOW USERS SEARCH TRIP PLACES
A. Requirements in search of trip places
This paper defines a trip goal as what a user wants to
experience in a vacation place. For example, a user wants to
relax her body and mind with a hot-spring or refresh her in
nature. In addition to that, the paper defines a trip preference as
what the user wants to experience in the trip to achieve her trip
goal. For example, she prefers outdoor hot springs to others,
or wants to view a deep-blue sea from a height to refresh in
nature. Before the trip, the user would enjoy imaging the way
she would experience when she visits a specific vacation place
to achieve her trip goal with her trip preference. The paper
refers to a trip image as what she envisions for experience
she would take in the trip. Suppose a user desires to relax her
mind and body, or to refresh her viewing deep-blue sea from
a height. She would search for a vacation place, imaging her
experience of them. Before a trip, a user would mostly plans
for the trip. She plans a trip according to conditions in the
trip such as trip goals, partners, their preference, and so on.
Imaging concrete situations in the trip under the conditions,
the user chooses a specific vacation place. Let us consider a
trip with intimate friends for a long time. The trip planner
comprehends well common preference between her and the
intimate friends. She would make a trip plan, choosing vacation
places which make both of her and the friends enjoy. When she
thinks those vacation places, she images herself and her friends
enjoying scenes and their conversation at those vacation places.
When she thinks that one of those vacation places is suitable
in this trip, she decides to go on a trip in that place. In this
way, the user enjoys the process of deciding a vacation place
imaging situations in the trip, as well as the trip itself. It is due
to the increase of her expectation for their trip by the imaging.
She gets excited, while she is making concrete trip images
along with the trip goals and trip preference. Sometimes, she
finds new trip goals or trip preferences which excite her,
because she attains unknown vacation information. It enhances
options in the trip, which makes the trip plan more pleasant
for her. On the other hand, if she fails to make her trip goals
and trip preference clear, she can never proceed to the step to
decide on an exciting vacation place, because she imagines no
concrete situation. Travel agencies would recommend vacation
places, interviewing travelers. In the interviews, they would
make unclear trip goals and trip preference to clear ones to
excite the travelers, in order to find good vacation places. One
of roles of travel agencies is to provide unknown vacation
information for customers, increasing their expectation for
their trip.
B. Searching for vacation places with Internet
We have conducted a questionnaire to survey user re-
quirement for time used for the search of vacation places, in
advance. In the questionnaire, users are asked whether they
prefer search in a short time if they can find suitable vacation
places. About 92% people respond “ yes ” in this question.
It reveals people want methods with which they can find
information to decide on vacation places in a short time and
without efforts. As one of those methods, a lot of people use
the Internet. According to the investigation in 2012 [1], 45.2%
people gather the vacation information using Internet, which
means the Internet is more frequently used than any other
major medias such as TVs, radios, and newspapers. One of
reasons which make the Internet the most popular tool is that
we can search for the travel information anytime and anywhere
with it. For example, if users ask travel agencies to make travel
plans, they must visit the agencies during their business hours.
In other words, they cannot make plans for trips if they have
no time in the business hours. Using the Internet, they can
search for travel information even in the midnight. On the other
hand, there is too much travel information on the Internet. A
user must select appropriate travel information so as to decide
on a vacation place she wants to visit. Keywords on vacation
places must be specified in the existing search engines, such
as Google [2] and major support sites for travel planning [3].
Suppose a user desires to visit hot springs she can visit within
3 hour driving from her home town. She specifies the keywords
such as the province names around her home town and ”hot
spring” in the search bar of Google. It provides many web
pages including the province names and keyword ”hot spring”
in their titles and contents. She looks for useful information
in order to plan her trip from the Web pages.
C. Problems in Internet search
In the Internet search with keywords, a user must embody
her trip goals and preference in advance, so that she can specify
them in the search bar. It means that she can search for suitable
vacation place using the Internet only if she has embodied
her trip goals and trip preference. The constraint brings about
harmful influences. Users with little knowledge on trips would
have difficulties to embody their goals and preference for trips.
The lack of embodied goals and preference prevents them
from expanding trip images. Without expanded trip images,
they cannot conceive good keywords to be specified in search
with the Internet. Consequently, they have little chance to find
suitable vacation places matching their trip images. On the
other hand, users with a lot of trip experiences have often fixed
their trip preference. When they search for vacation places
20Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
28 / 79
in the Internet, they are likely to specify same keywords in
all cases. They are disappointed with stereotyped trip images
brought by search results. They would not be excited in
conventional events which they experienced in different places.
They experience nothing novel. They fall into a condition that
there is no places which they want to visit although they want
to go on trips. There is a common issue in both of users lacking
trip experiences and ones abundant in trip experiences. They
cannot figure out clear trip goals and preference enough to
make them exited, when they make trip plans. In order to
make search of vacation places with the Internet beneficial to
them, it is inevitable for them to embody their trip goals and
preference.
D. Related works
There are existing studies to recommend vacation places
to users using the Internet [4][5][6][7]. As it is described
in the previous section, users have to embody their goals
and preference of trips so as to express them as keywords.
In order to solve this problem, the works in [8], [9] and
[10] address to recommend suitable vacation places matching
their trip goals and preference. These works use location
information of pictures the users have taken in vacation places
as their travel histories. The works recommend vacation places
through the user-based collaborative filtering from the travel
histories. However, they leave 2 issues unsolved. First, they
cannot provide any vacation places matching new trip goals
or preference, because they use travel histories. Users cannot
be excited when they recommend the users vacation places
similar to what the users have been to. Second, users get
excited when they are going to experience something matching
their latent trip goals and preference. However, it is not easy
to embody latent trip goals and preference so that they can
express them as keywords. In order to solve these issues, we
need to let users get aware of their unknown trip goals and
preference they have never experienced. In addition, we need
to reflect their unknown trip goals or preference to the result
of recommendation. Furthermore, according to Section II-A,
it is important for all users to identify their trip goals and trip
preference, in a short time and with little effort.
III. METHOD TO EMBODY TRIP GOAL AND PREFERENCE
A. Categories and tags
The paper proposes a method to embody trip goals and
preference of users from their browsing logs of trip place
photographs on a web site. When the users see photographs
of vacation places, users can momentarily judge whether they
want to visit or not there, because photographs give users
more intuitive impressions than texts. In addition, photographs
represent vacation places as they are. Users can expand easily
their trip images through photographs of vacation places.
Photographs express moods, which are difficult for texts to
convey. Consequently, users can expand their trip image from
photographs more concretely than text. The paper assumes
that a photograph presented to users on a web site has one
big object and background, as depicted in Figure 1. When
users looking for vacation places see photographs of specific
sightseeing points, they immediately judge whether vacation
places in the photographs match with their trip goal or not. If
they match, the method assumes they want to enjoy the mood
big objects and backgrounds in the photographs bring. This
Fig. 1. Photograph using proposed method
paper refers to the mood as a category. It is expressed with
a phrase such as ”hot springs” and ”landscape”. The category
is used to estimate a trip goal of an user. In addition to that,
this paper refers to the object shown in the photograph as its
core as a tag. The tag is also expressed with a phrase such as
”outdoor bath” and ”sunset”. The tag plays an important role to
estimate trip preference of the user. When a user searches for
vacation places using photographs, it is easy for her to image
herself in the vacation place. Imagining herself in the vacation
place, she considers whether the background and the object
in the photograph suit her trip preference. In this process, she
embodies her unclear trip preference, seeing the object in the
photograph.
Photographs used in this study have more than one category
and tag. For example, the photograph in Figure 1 has the
categories, ”nature” and ”Buddhist temple”, and the tags,
”tinted autumnal leaves” and ”temple”. Various categories and
tags are chosen to represent trip goals and preference common
to many users. We prepare so many photographs that they
should cover all common trip goals and preference without
loss. We make correspondence of each photograph to some of
trip goals and preference. Users can embody their unknown
trip goal and preference when they see the photograph in the
process of imagining their trips.
B. Embodying trip goal and preference from photographs
The method makes trip goals and preference of users clear
so that vacation places can be easily searched with exciting
search engines on the Internet. It lets users embody their trip
goals and preference, which would be used as keywords for
search engines. Users are supposed to specify the keywords for
search engines to find vacation places suitable for the users.
Figure 2 illustrates the overview of the method. In the method,
users increase points of photographs which contain her favorite
combination of objects and backgrounds. The increment is
referred to as“ evaluation”by users. The points are provided
for categories and tags associated with the photographs. In
the screen where many trip photographs arranged in a random
order, the method lets users evaluate her favorite ones. The trip
photographs are arranged on the screen, according to the order
calculated with Equation (1). p
m
is the probability category m
is selected. E
t
is the entropy of time point t.
E
t
= −
∑
m
p
m
log
2
p
m
(1)
Through the user evaluation of photographs, the method
estimates their trip goals and preference. As Equation (1)
21Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
29 / 79
Fig. 2. Overview of method
shows, trip photographs on one screen are arranged so that the
entropy in the screen is maximized in terms of trip goal and
preference. When users choose specific ones from randomly
arranged photographs, they are biased against the particular
trip goal and preference represented with the photographs.
Suppose a specific category or tag reach is evaluated more than
a predefined threshold times by a user. The user is estimated to
have interests in a trip goal or preference corresponding to the
category or the tag. At that time, a new screen appears. The
screen is filled up with new photographs corresponding to the
category or the tag chosen in the previous screen. The method
lets the user evaluate photographs in the new screen, again. It
aims at estimating other trip goals or preference, assuming
the uses have the chosen trip goals and preference. When
the number of estimated trip goals and preference reach to
predefined one, the method finishes evaluation of photographs
by user, considering it has already gather trip goals and
preference sufficiently. After that, it passes estimated trip goals
and preference as keywords of search engines to recommend
vacation place.
C. Evaluation with enlargement and selection
The method estimates trip goals and preference of a user
by evaluating photographs. Figure 3 shows the screen for eval-
uation with photographs. From the screen, a user is assumed
to enlarge or select photographs which suit her trip goals and
preference.
A photograph is enlarged when its main object pleases her,
as depicted in figure 3. Since the size of photographs for the
evaluation is 64 pixels in both of the length and width, the
user would desire to make their image clearer with a simple
way, when they focus on them. A user can enlarge photographs
when the mouse pointer hovers over them more than 1 second.
When the user enlarges the photograph, the method adds a
point to the categories the photograph corresponds to, because
the method regards she likes the mood it presents.
The photograph is selected when the enlarged photograph
is clicked. When the user judges the photograph suits her trip
preference after her enlargement, she selects it to confirm its
detail. The method considers the user takes the selection action,
because she has strong interests in the photograph after the
enlargement. The method regards she tries to confirm its detail
Fig. 3. Screen of evaluation
with the selection, when she has judged its object suits her trip
image well.
Suppose a user enlarges and selects the photograph having
a hot-spring category and an outdoor bath tag. The method
estimates that she is pleased with the mood the photograph
presents. It adds a point to the hot-spring category. At the same
time, the selection makes the method estimate that photograph
suits her trip preference. The Outdoor bath tag gains a point.
D. Recommendation of vacation places
The trip goals and preference of the user are extracted as
categories and tags represented with phrases. Using extracted
phrases as keyword, the method searches for vacation places
with a major travel support site [3] for travel planning. The
site provides vacation places associated with the keywords in
their information and reviews from other travelers. The method
recommends the top 5 vacation places in the recommendation
ranking. It recommends more than one vacation places because
the user can choose geographically suitable places for her trip.
Suppose the categories and tags finally estimated as trip goals
and preference of the user are “hot spring”, “outdoor bath”,
“sunset” and “sea” through the evaluation of photographs. The
method searches for vacation places, presenting the estimating
phrases as keywords. As a result of the search based on the
keywords, some hot spring resorts are presented such as ones
allowing traveler to enjoy sunset from the outdoor bath.
IV. USABILITY OF RECOMMENDATION USING
PHOTOGRAPHS
A. Threshold for evaluation times
We have developed a Web based system to evaluate
the usability of the method. The system presents numerous
photographs using several screens. When a user evaluates
a specific category or a specific tag more than predefined
times, another screen appears, filled with new photographs.
The thresholds for the evaluation times for categories and tags
are θ
c
and θ
t
, respectively. We have conducted a preliminary
experiment in order to decide on the thresholds. The prelim-
inary experiment examines the combination of the 2 kinds
of thresholds which brings the largest satisfaction of users
for vacation places. Let us refer to the number of enlarging
or selecting photographs associated with a specific category
or a specific tag as the number of evaluation times. Table I
22Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
30 / 79
TABLE I. COMBINATION OF THE NUMBER OF EVALUATION
A B C D
Number of evaluation for category 8 10 12 14
Number of evaluation for tag 4 5 6 7
TABLE II. LIST OF CATEGORIES AND TAGS (1)
landscape nature hot springs
seaside tinted autumnal leaves outdoor bath
native province cherry blossoms indoor bath
prospect bamboo forest hot spring district
river pine-covered area public bath
star sky valley
night view forest
shows 4 kinds of combinations of the number of evaluation
times for categories and tags examined in the preliminary
experiment. 100 Vietnamese university students from 19 years
old to 20 years old have joined to the experiment as test
subjects. The 100 test subjects are divided into 4 groups, each
of them consists of 25 people. Let us refer to them such
as “A”, “B”, “C” and “D”. In the preliminary experiment,
subjects use the web site, imaging the place which they want
to visit in Japan. They choose their favorite vacation places
among recommended ones. They express their satisfaction
degrees for each vacation place recommended as a final result
in a questionnaire. Since combination C brings the highest
satisfaction, the thresholds for the evaluation times to present
a new screen are θ
c
= 12，θ
t
= 6.
B. Experiment to verify usability of photographs
We have conducted a main experiment to verify the us-
ability of this method, under the following condition. When
the number of estimated trip goals and preference reach to
predefined thresholds for trip goal, θ
p
, and for trip preference,
θ
o
, the method finishes evaluation by a specific subject. In the
main experiment, θ
p
= 3 and θ
o
= 2. Furthermore, the method
finishes the evaluation in case a subject evaluates photographs
more than the predefined maximum number. In the main
experiment, the maximum number of the evaluation is 50.
Table II and Table III show the categories and tags used in this
experiment. The categories and tags are determined referring
to the items of trip genre in a travel support site for travel
planning [3] so that any prejudice affects the arbitrariness
of the experiment. To prepare trip photographs used in the
experiment, the names of categories and tags in the Table
II and III are specified as keywords in Google visual search
[11]. From photographs the site presents, we manually select
photographs suitable for trip photographs from viewpoints of
comprehensibleness.
There are 4 points to be verified in this experiment. The
first point is the ability of trip photographs to allow users
embody their trip goals and preference. The second one is the
satisfaction degree of users for vacation places recommended
based on the embodiment of their trip goals and preference.
Third, we have examined whether users can search for vacation
places matching their trip goals and preference in a short time.
Finally, we verify the method enables users to search for more
easily vacation places, compared with the major support sites
TABLE III. LIST OF CATEGORIES AND TAGS (2)
event sport Buddhist temple
carnival camp temple
illuminations water sports shrine
the Star Festival mountain climbing
tradition ski slope
sea bathing
TABLE IV. LIST OF CATEGORIES AND TAGS (3)
leisure spots vehicle history Japanese-style hotel
amusement park sightseeing boat Japanese castle seafood
aquarium ropeway historic spot food of the mountain
zoo truck historic building tea lunch
botanical gardens old highway atmosphere
shopping garden European
art gallery
pasture
for travel planning. The test subjects are 18 male students and
7 female students of universities in Japan. Their ages range
from 18 years old to 26 years old. The experiment compares
the proposed method with the way with which people generally
search for vacation places with search engines through the
Internet. The paper refers it to as the conventional way. This
experiment supposes 2 situations. The first one supposes a user
searches a spring trip in a place where she and her friends have
never been. In the second one, a user searches for vacation
places which please her close friend or her boyfriend in their
winter vacation. In this experiment, 25 test subjects are divided
into 2 groups: A and B. The grouping aims to reduce the
influence of test subjects which have variance in their ability
to image the specified situation easily. The experiment lets one
group apply the proposed method to the spring situation and
the conventional method to the winter situation, while the other
searches for vacation places for the spring situation with the
conventional method and for winter situation with the proposed
method. This experiment lets each of test subjects search for
a single vacation place which best suits for an image the test
subject has for each situation.
In the experiment, first, a user searches for one vacation
place which satisfies her best for the specified situation, using
the conventional method. Second, using the proposed method,
the user chooses one vacation place which best suits for the
other situation. Finally, in regard to the vacation places they
find out in the process above, we have asked several questions
to the user. After the proposed method estimates the trip
goals and preference of the user, we have prepared 3 sets
of keywords; the first one consists of only the trip goals,
the second includes only the trip preference, and the last is
composed of both of the trip goals and the trip preference. For
each of 3 keyword sets, we have collected 5 vacation places
at most. Since the number of collected vacation places varies
with keyword sets, the total number of recommended vacation
places ranges from 5 to 15. Each test subject determines a
single vacation place he or she wants to visit most.
C. Embodiment of trip goals and preference
We examined whether users can embody their trip goals
and preference with trip photographs. In the questionnaire
23Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
31 / 79
Fig. 4. User satisfaction
TABLE V. Average and dispersion of working time
Conventional method Proposed method
Average 35.32 7.045
Dispersion 387.25 12.39
asking “Can you embody your trip image with the trip pho-
tographs?”, about 92% users said “yes”. It showed users were
able to embody their trip image. It seemed that users enlarged
or selected trip photographs to see the details of what were
presented in the photographs. They examined the details to
see whether the photographs met their goals and preference.
Operations of enlarging or selecting photographs implied trip
goals and preference of users. Furthermore, in interviews after
the experiment, some users uttered that the photographs made
them aware trip preference they had never found. As a result,
users were able to find out their latent trip preference from the
photographs.
D. Satisfaction for vacation places
Let us see whether users were satisfied with vacation
places recommended based on their trip goals and preference.
According to the results of the experiment, about 64% of
users were satisfied with recommended vacation places. Figure
4 compares the number of users satisfied with the proposed
method with that of users satisfied with the conventional
method for the same situations. The comparison of the 2 results
indicates the proposed method recommended vacation places
with the same level of satisfaction as the conventional method.
E. Time for searching
Since most users need a recommendation method which
presents vacation places with few operations, it is important
whether users can search for vacation places corresponding
to their trips goals and preference in a short time. Table V
compares the proposed method with the conventional method
in terms of the working time, and its mean along with the
deviation, respectively. Users took 7 minutes in average to find
out favorite vacation places with the proposed one, while 35
minutes in average with the conventional one. Moreover, all of
users were able to find out the vacation places in shorter time
with the proposed one than the conventional one. Together
with the result of the previous section, the proposed method
contributes to finding as much satisfying vacation places as the
conventional method in a shorter time.
F. Effort for searching
It makes no sense if users need lots of effort to search
for vacation places using this method. There are the results
of the questionnaire asking the following 2 points. The first
question is “which method makes you go into deep thought,
the conventional method or the proposed method?” The second
is “which do you think impose more bothersome operations,
the conventional method or the proposed method?” In the first
question, about 72% users said “the conventional method”.
In the second question, about 76% users said “the conven-
tional method”. In regard to the result, we have tested any
difference of this method from the conventional method with
the chi-square test. The null hypothesis was that there was no
difference between the conventional method and the proposed
method, while the alternative hypothesis was that there was
a difference between them. Table VI and VII show the test
results. Regarding to the first question, there was a difference
at the significant level of 5%. There was a significant difference
in at the 1% level as for the second question. From both
viewpoints, the proposed method allowed users to search for
easily vacation places than the conventional method.
TABLE VI. RESLUT OF FIRST QUESTION
Answer
Chi-square 4.84
Flexibility 1
p-value 0.0278069
TABLE VII. RESULT OF SECOND QUESTION
Answer
Chi-square 6.76
Flexibility 1
p-value 0.009322376
V. DISCUSSION
A. Improving precision of estimation
Let us consider to improve the precision for the estimation
of trip goals and preference of users. In the questionnaire to
be answered in 5 grades, about 36 percent of users answered
“ normal ”or“ a weak dissatisfaction ”with vacation places
the proposed method recommends. As reasons of the dissatis-
faction, some users mentioned the followings.
1) “ It does not seem photographs I have chosen are not
reflected to determine vacation place candidates. ”
2) “ Though I have chosen many food photographs, the
method mainly presents photographs of ruins or red
leaves of autumn in the next screen. ”
The dissatisfaction may attribute to low quality in estimating
trip goals and preference of users using categories and tags
associated with photographs.
Reason 1 implies a problem on the way to select new
photographs to be presented for users from their operations
on previous photographs. New photographs are selected using
categories and tags attached to previous ones. It is suspected
categories and tags are determined based on the arbitrariness
of the implementers of the proposed method. Suppose a user
prefers a photograph because she likes the sky presented in
it, while tags indicating a temple and red leaves are attached
24Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
32 / 79
to it. Though the method values tags indicating a temple and
red leaves, it never picks up the trip preference of the user
directing the sky. In this experiment, the implementers have
determined manually categories and tags, which allows the
arbitrariness of the implementers to be reflected strongly on
them. Consequently, it is conceivable trip goals and preference
are estimated using photographs whose categories and tags
are biased. It prevents the method from estimating proper trip
goals and preference users have. We need an objective way to
determine relevant categories and tags.
Reason 2 notifies us of unfairness in the evaluation of
photographs. The method evaluates all categories and tags
with a constant weight, when a user enlarges or selects a
photograph. Objects on photographs have large variety in their
size. Depending on the size, they provide different impressions.
For example, if a user looks at a photograph with cherry blos-
soms expanding, she necessarily focus on the cherry blossoms.
However, if a photograph shows cherry blossoms in small size
in front of big Mt.Fuji, a user may not always focus on cherry
blossoms. The evaluation with constant weights is unfair for
photographs containing objects of various size. One way for
the improvement is to reflect the size ratio of objects to weight
to evaluate photographs.
Solutions on these problems would improve the precision
in the estimation of trip goals and preference of users.
B. Improving satisfaction for vacation places
Some users show dissatisfaction for recommended vacation
places, because the method presents only places they have
already visited, or the method recommends vacation places
they cannot visit. A main reason for the dissatisfaction is lack
of consideration of travel histories and hometowns of users.
Without the consideration, the method may recommend vaca-
tion places users have been to, or cannot visit geographically.
The method can avoid the problems, taking travel histories
and hometowns of users into consideration in advance. It
needs to construct a data base reflecting travel histories and
hometowns of users. The data base should classify vacation
places associated with categories and tags into geographical
areas to calculate the accessibility of users. It should also
register vacation place each user has been to in advance. Rec-
ommendation of vacation places using the data base satisfies
more users.
VI. CONCLUSION
This paper has proposed a method to embody user require-
ments for trips. The method estimates trip goals and preference
of users who enlarge and select trip photographs during they
are making trip plans. After the estimation of the trip goals
and preference, the method specifies them as keywords in a
major support site for travel planning, to get recommends on
vacation places which suit for trip goal and preference of each
user. An experiment has proved about 64% users are satisfied
with vacation places recommend using this method, which is as
reliable as conventional methods. All of users can materially
shorten the time to decide on their vacation places with the
proposed method than with the conventional one. Users can
decide on trip plans suitable for them in a short time with
the method. We have also discussed objective categories and
tags, as well as construction of a data base to enhance user
satisfaction as our future works.
REFERENCES
[1] Ministry of Public Management, Home Affairs, Posts and Telecom-
munications (2012) White paper Information and Communications in
Japan: http://www.soumu.go.jp/johotsusintokei/whitepaper/ja/h24/html/
nc123340.html [retrieved: 12, 2014]
[2] Google: http://www.google.co.jp/ [retrieved: 12, 2014]
[3] Jaran saported by Recruit: http://www.jalan.net/ [retrived: 12, 2014]
[4] Vincent. W. Zheng, et al. ”Towards mobile intelligence: Learning
from GPS history data for collaborative recommendation.” Artificial
Intelligence 184 2012, pp. 17-37.
[5] Ye. Mao, et al. ”Exploiting geographical influence for collaborative
point-of-interest recommendation.” Proceedings of the 34th interna-
tional ACM SIGIR conference on Research and development in In-
formation Retrieval. ACM, 2011, pp. 325-334.
[6] Michele. A. Burton, et al. ”Crowdsourcing subjective fashion advice
using VizWiz: challenges and opportunities.” Proceedings of the 14th
international ACM SIGACCESS conference on Computers and acces-
sibility. ACM, 2012, pp. 135-142.
[7] Joan. Borrs, et al. ”Intelligent tourism recommender systems: A survey.”
Expert Systems with Applications 41.16 2014, pp. 7370-7389.
[8] Marco. Mamei, et al. ”Automatic Analysis of Geotagged Photos for
Intelligent Tourist Services.” Intelligent Environments. 2010, pp. 146-
151.
[9] Maarten. Clements, et al. ”Using flickr geotags to predict user travel be-
haviour.” Proceedings of the 33rd international ACM SIGIR conference
on Research and development in information retrieval. ACM, 2010, pp.
851-852.
[10] Adrian. Popescu. and Gregory. Grefenstette. ”Mining social media to
create personalized recommendations for tourist visits.” Proceedings of
the 2nd International Conference on Computing for Geospatial Research
& Applications. ACM, 2011, pp. 37.
[11] Google visual search: https://www.google.co.jp/imghp?hl=ja&tab=wi
[retrived: 12, 2014]
25Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
33 / 79
Estimating Consumer Inclination for Agricultural Products
from Web Browsing History
Riki Tatsuta, Sunao Nakanishi
Graduate School of Information Science and Engineering
Ritsumeikan University, Shiga, Japan
email:{tatsuta, sunao}@de.is.ritsumei.ac.jp
Yusuke Kajiwara, Hiromitsu Shimakawa
College of Information Science and Engineering
Ritsumeikan University, Shiga, Japan
email:{kajiwara, simakawa}@de.is.ritsumei.ac.jp
Abstract—In recent years, more and more e-commerce sites
sell agricultural products. It should be noted that users have
preference of taste. They also have intention on what kinds of
agricultural products they get. They consider them at the time
of purchase of agricultural products in recommendation on e-
commerce sites. There are a number of studies on the automatic
extraction of preference of taste that users have from contexts on
the web. However, there has been no study that tries automatic
extraction of user intentions to purchase agricultural products
on e-shopping sites. E-shopping sites without consideration of
user intentions are likely to recommend agricultural products
far from what users want to buy. Confused users will never visit
such e-shopping site, again. We propose a model which estimates
intention level of users from web contents of e-commerce sites
and web browsing history of users. The proposed method refers
to the degree of user interest about a topic calculated from LDA
as the topic attention level. Based on the multivariate logistic
regression model, the method constructs an intention level model
from the topic attention level. An experiment suggests the method
can calculate the safety intention level, focusing on price of
agricultural products on web pages.
Keywords–recommendation; inclination; intention; preference;
agriculture; e-commerce site; browsing history.
I. INTRODUCTION
More and more e-commerce sites sell agricultural products.
Sale of agricultural products on the Internet has several good
effects. Farmers can increase their sales. They are also moti-
vated through direct communications with consumers. Supply
of high-quality agricultural products provides them with good
brand impression [1]. E-commerce brings convenience of
direct delivery of agricultural products to consumer houses.
The elderly households are too weak to go out shopping
food. Households who have small children have no time to
go out for shopping. They are eager to buy food with Internet
[2]. On the other hand, farmers need efforts to make more
consumers visit their own sites to make success in e-commerce
selling agricultural products. Their e-commerce sites present
information which motivates consumers to purchase their
products. In addition to reviews, prices, and quantity like
general e-commerce sites, they have information on varieties
and palatability of products. They also present cultivation
methods to explain their cultivation policy [3].
Information motivating consumers to buy agricultural prod-
ucts varies with preference of consumers. The preference
is determined by various factors. For example, preference
on cooking with agricultural products ranges from foodstuff,
menu, mouthfeel, taste, seasoning and flavor [4]. Apart from
these, consumers have intentions on their peculiar ideas such
as safety and costs [5]. Some consumers give a higher priority
to safety and palatability than to prices, while others mind
prices rather than appearance. A combination of preference
and intentions of each consumer forms specific inclination
which regulates purchase behavior of the consumer. Farmers
need marketing activities to target consumers whose inclination
meets their cultivation policy. However, Japanese farmers en-
gaging in highly intensive agriculture cannot afford to consider
marketing activities because they are busy with cultivating
many kinds of crops. One possible solution to support attract-
ing consumers is Search Engine Optimization (SEO), which
makes the site appear more frequently in result list of Internet
search engines. However, the solutions cannot support the site
builders should make consumers interested in the sites so that
the consumers can buy agricultural products buy according to
their inclination. From the viewpoints of consumers, they want
to buy agricultural products from farmers whose cultivation
policy meets their own inclination. For example, consumers
with safety intentions would read politely safety explanations
of products in the e-commerce sites, because they want to
purchase safe products as possible, as far as their price
permits. They want to get foods free from pesticides. Trading
on the current e-commerce never considers coincidence of
farmer cultivation policy with inclination of consumers, when
customers search farmers optimal to them. In fact, consumers
must manually look for agricultural products fitting their needs
from a lot of e-commerce sites, struggling with search engines.
However, search engines disappoint them many times, intro-
ducing sites which handle mismatching agricultural products.
This paper proposes a method to estimate inclination of
consumers, in order to recommend appropriate e-commerce
sites to them when they search farmers. The method utilizes
Web contents on the e-commerce sites and browsing history
of customers. The method figures out the distribution of
topics on farmer cultivation policies from a specific Web page
using the Latent Dirichlet Allocation (LDA). The paper refers
the product of the topic distribution and the browsing time
of the web pages as the topic attention level. The method
constructs an inclination level model using the topic attention
level as predictor variables, based on the multivariate logistic
regression. An experiment result suggests the method can
calculate the safety inclination level.
This paper is organized as follows. Section II discusses the
detail of inclination. The related works are also explained in
Section II. Section III describes the outline of the proposed
26Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
34 / 79
considered recommendation system. The proposed method to
estimate inclination is also described in Section III. Section
IV presents the results of experiments and evaluation. Section
V presents the discussion on improvement of the proposed
method. Finally, Section VI concludes this papers.
II. PRODUCT INFORMATION MATCHING INCLINATION
Consumers have preference on agricultural products, such
as taste, appearance, flavor and so on. The preference is said
to differ by age, generation, gender, regions and occupations
[6]. In addition, consumers would stick to their peculiar ideas
such as safety and costs, when they choose products to buy.
Consumers generally determine products to buy with their
preference and intentions. The paper refers the combination
of preference and intentions of each consumer as inclination.
It usually regulates purchase behavior of the consumer.
There are studies on the estimation of consumer inclination
[7][8]. For the estimation, the studies oblige consumers to fill
up questionnaires and profile in free description. They analyze
results with text mining technologies to find user inclina-
tion for agricultural products. However, consumer inclination
changes depending on fashion and trend. For example, users
minding safety, which means products are free from harmful
substance to eat, have greatly increased in Japan because of
the Fukushima nuclear accident. It is necessary to successively
examine consumer inclination because the thinking ways of
consumers vary from day to day. It is an unendurable for
consumers to fill up questionnaires and profiles many times. A
new method is required to extract current inclination of each
consumer without bothering the consumer.
III. RECOMMENDING AGRICULTURAL PRODUCTS
MATCHING INCLINATION
We aim at detecting user inclinations from Web browsing
log before they decide which farming products they buy. It
does not make any sense to analyze huge data obtained after
they decide to buy. The log of URLs in Web browsing of
users is also insufficient to achieve the aim above. We need to
analyze the contents to see what topics in the pages stimulate
the users. The LDA on the contents of Web pages they are
interested in enables us to identify their topics.
A. Outline of recommendation with extracted inclination
To extract inclination of consumers at the time, we use their
browsing history while they look for their favorite agricultural
products. We extract consumer inclination from Web pages
the consumer visits while he chooses agricultural products
in e-commerce sites. E-commerce sites present a variety of
Web pages. In the web pages, there are contents which
provide stimuli to consumers, because the contents match their
inclination. Inclination of each consumer is likely to appear
when he selects products. It seems that Web pages vary with
consumers because of their inclination. For example, it may
be supposed consumers who put importance to safety purchase
domestic or organic tomatoes. In this case, they are supposed
to investigate description on agricultural chemicals or product
places. On the other hand, consumers preferring economy to
safety purchases foreign tomatoes, because of low price. In this
case, consumers would browse pages of agricultural products
after they sort the pages in the decreasing order of the price. It
may be suspected that browsing time of Web pages gets long if
Browses the web page
Recommendation
browsing pages and time
䐟Getting browsing
history
䐡Estimating user’s
intention level
䐠Calculating
intention model
browsing history
user’s intention level
intention model
safety domestic
locality
economy
gourmet
Figure 1. System configuration diagram
they have stimulated consumers. In the proposed method, we
estimate consumer inclination from topics described in each
Web page and the browsing time to read the Web page. In this
work, we focus on inclinations: safety, domestic production,
locality, economy, and gourmet from the survey results of
inclinations on foods [6] and the studies of selection criterion
at purchase of agricultural products [9]. As Figure 1 shows, the
paper proposes a method to build a model to estimate consumer
inclination from browsing history of the consumer to provide
him with agricultural product information. At first, the method
extracts browsing pages and browsing time from a browsing
history when the consumer chooses agricultural products on
e-commerce sites. Second, it calculates the degree of how
much the consumer has each inclination. It refers to the degree
as the inclination level. Finally, the method recommends the
agricultural product information which suits his inclination.
There are various agricultural products. For example, or-
ganic tomatoes, fruit tomatoes, cherry tomatoes are classified
into the category of tomatoes. These categories are referred
to as product categories in the paper. Though consumers
searching products visit various sites, the method values pages
for a specific product category as item pages, rather than pages
showing lists of item pages, which are referred to as search
pages. Pages irrelevant to e-commerce sites such as Google
search and Wikipedia are referred to as other pages in the
paper. Item pages are classified into product categories, based
on words on their titles and contents. From browsing history
of each consumer, item pages are extracted for each product
category. As Figure 2 shows, the proposed method builds a
model to estimate the inclination of each consumer as follows.
1) The item pages are regarded as a set of words
according to the Bag-of-words method [10]. The
topic distribution of the item pages is figured out
with the LDA. The proposed method calculates the
topic attention level with the product of the topic
distribution and the browsing time of each item page.
2) The method selects important variables to calculate
the inclination level of the consumer.
3) Using the inclination level, the method constructs an
inclination model for each consumer.
B. Weighted topic distribution
Through a morphologically analysis, the proposed method
divides a set of sentences in item pages into words. It makes
27Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
35 / 79
Browsing pages and time
LDA
Topic distribution
Topic attention level
Word group
Intention model
Multivariate logistic
regression
Intention of user
safety
economy
gourmet
1
0
1
Object variable
Bag of words
Questionnaire
Extraction
Explanatory variables
Comparison
Weighting
Figure 2. Proposed method
bag-of-words from nouns, adjectives and verbs which seem
to be related to inclination. Using the LDA [11], the method
acquires combinations of topic distribution and its word group
for each page. Though words in the item page are relate to
various kinds of inclination, the LDA allows to get topics
classified according to each inclination.
The proposed method uses browsing time to determine
whether the item page is related to a specific inclination or
not. The previous work [12] has clarified a relationship of the
browsing time of a Web page to the degree of user interest.
LDA is a model which assumes the document is approved from
two or more topics. The degree of the interest to a document
is that to the topic on the document. The degree of the interest
to the topic of the page are related with user’s browsing time.
The words which relate to several kinds of intention may be
included in the e-commerce sites selling agricultural products.
However, if each page is weighted by browsing time, user’s
focus point of words of the page is not considered.
Words which relate to the intention should be classified
in one page. This technique subdivides words by a topic
distribution of each item page, which enables the acquisition
of the degree of each of the user’s intentions. Equation (1)
shows topic distribution the vectors T
cp
.
T
cp
= [
T
cp1
· · · T
cpk
· · · T
cpK
]
T
(1)
(1 ≤ c ≤ C), (1 ≤ k ≤ K), (1 ≤ p ≤ P
c
)
where k , c, and p represent a specific topic, a specific product
category, and a specific page, respectively. K and C are the
number of topics and product categories, respectively. P
c
is the
number of item pages which are contained in the c-th product
category. T
c
pk is the k-th component of topic distribution of
the p-th page of the c-th product category. Equation (2) shows
vector I
c
. Each component of it is a sum of T
c
p weighted by
the browsing time of each page in each product category.
I
c
= [
I
c1
· · · I
ck
· · · I
cK
]
T
=
P
c
∑
p=1
(T
cp
)d
cp
(2)
I
ck
is the weighted vector component relating to the k-th topic
of the c-th product category. d
cp
is the browsing time of the
p-th page belonging to the c-th product category. The more
pages belonging to the same commodity category the consumer
browses, the more I
ck
grows. It is necessary to normalize I
ck
,
because the browsing time and the browsing count vary with
consumers. Equation (3) shows S(I
ck
) which is calculated
with normalized I
ck
.
S( I
ck
) =
I
ck
|I
c
|
(3)
In this work, we refer to S(I
ck
), which is extracted from each
product category as the topic attention level.
C. Selection of important variables
In Section III-C, we account for the explanatory variables
of the making model as shown in Section III-D. The topic with
difference of topic attention by the presence of the intention
is extracted. Equation (4) shows G{n
ck
}, the set of the topic
attention level of each consumer.
G{n
ck
} := { S(I
ck
)
1
, · · · , S(I
ck
)
u
, · · · , S(I
ck
)
U
} (4)
(1 ≤ u ≤ U )
Whole users are divided into 2 groups: one with a specific
inclination and the other without it. In Equation (5) and (6),
X{n
ck
} and Y {n
ck
} show consumers with the n-th inclination
and without it, respectively.
X{n
ck
} := {G{n
ck
}|P ({G
ck
})} (5)
Y {n
ck
} := {G{n
ck
}|Q({G
ck
})} (6)
P is the condition that the consumer does not has the inclina-
tion. Q is the condition that the consumer has the intention. n
and N corresponds to the n-th inclination and the number of it.
n
ck
is the k-th topic of the c-th product category to correspond
to the n-th inclination. The method aims to build a model
to estimate inclination of unknown consumers. It assumes
inclination of several consumers is acquired beforehand by
questionnaires or interviews. Though it obliges some efforts on
28Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
36 / 79
some consumers, there is no burden on unknown consumers.
The method calculates difference of the topic attention level of
the group with the inclination and that without the inclination.
Let us focus on the difference of the topic attention levels
of the two groups. A large difference in a topic indicates the
group with the inclination pays their attention on the topic,
while the other group pays no attention on it. To examine the
difference of the topic attention level, the method uses median
values resistant to noises. Equation (7) and (8) calculate the
median values of the group with the inclination and the group
without it.
MedianX(n
ck
) =
{
x
(a+1)/2
, if a is odd.
1
2
(x
a/2
+ x
a/2+1
, if a is even.
(7)
MedianY (n
ck
) =
{
y
(a+1)/2
, if a is odd.
1
2
(y
a/2
+ y
a/2+1
, if a is even.
(8)
Equation (9) shows dif f (n
ck
) which is the difference of
MedianX(n
ck
) from MedianY (n
ck
).
diff (n
ck
) = MedianX(n
ck
) − M edianY (n
ck
) (9)
Equation (10) shows Diff
n
which is the set of the difference
of the median for the n-th inclination.
Diff
n
:= {diff (n
11
), · · · , dif f(n
CK
)} (10)
It seems that the median with great difference is an important
value to characterize a group of users who do not have
intention and a group of users who have intention. This method
picks out top R of large median difference in a positive
direction as High
n
and lowest R of large median difference in
a negative direction as Low
n
in Dif f
n
. Topics corresponding
to High
n
is focused by the group with the n-th inclination,
but neglected by the group without it. Topics corresponding to
Low
n
mean the opposite case.
D. Construction of inclination model
The proposed method builds a model to estimate the
consumer intention level. The explanatory variable necessary
for the model is a topic attention level which belongs to High
n
and Low
n
in III-C. The object variable is the inclination of
each consumer. The method refers to it as an inclination level
model. The method applies the multivariate logistic regression
model, to make a good inclination level model. To improve the
performance of the model, the method excludes explanatory
variables belonging to High
n
and Low
n
in a stepwise way.
The intention model distinguishes presence of a specific
inclination of an unknown consumer from his topic attention
level. The topics on each Web page are obtainable beforehand,
applying the LDA to the contents on e-commerce sites in
advance. What is required for the on-line calculation for the
inclination level of the consumer is only his browsing time.
IV. EXPERIMENT
We experimented to verify the utility of the proposed
method. The experiment aims at
• validation of efficiency of difference that user browses
topic by presence of the intention of the user, and
• verification of validity of intention level model.
A. Experiment method
We experimented about the purchase of agricultural prod-
ucts using e-commerce sites. The research participants were
17. We acquired the browsing history of the research par-
ticipant from beginning to the end of the experiment. We
had to make the research participants unaware of the costs
because that would have influenced their selection. We laid
out a situation as follows “I want to celebrate grandmother’s
sixty-first birthday with about 10 relatives. My mother said that
I should cook for the grandmother. You don’t need to think
about money, because I’ll pay. I thought that I will try to cook
in grandmother’s house, but there is no supermarket nearby.
So, I decided to prepare foodstuff using e-commerce sites.”.
In addition, the grandmother said “I want to eat home-made
dish because I dislike ready-made one. I want you to buy rice
you recommend.”. Below are the proposed dishes: salad, rice
cooked with matsutake mashroom, boiled dishes, and meat
dish. The material necessary for dish is shown as follows:
tomato, cucumber, matutake, rice, radish, taro, chicken, and
meat. We presented the amount of the foodstuff necessary for
the dish and the recipe of the dish to research participants. We
specified e-commerce sites to buy agricultural products. We
permitted them to use retrieval engines on Internet to gather
information. Each of the user’s intentions was acquired in the
questionnaire. We made a questionnaire, creating an associ-
ation between the information that the research participants
are interested in e-commerce site and each intention referring
to result investigation of intention [6]. It seems that the user
who has gourmet intention takes an interest in palatability. For
example, information is presented on e-commerce sites such
as ”sugar concentrations are higher and sweeter than another”,
”taste becomes better than another by the cultivation method of
original” and ”agricultural products harvesting in the morning
are very fresh”. The user who has a gourmet intention is
attracted by the deliciousness or the sugar concentrations. It is
seems that the user who has an economy intention is attracted
by low price or large quantity. For example, information
about price and quantity are presented on e-commerce sites.
Therefore, the user who has an economy intention will browse
a page that contains a price or quantity. However, we only
asked research participants about whether or not they are
attracted by price because of the quantity of agricultural
products that they should buy were specified. Questionnaire
items for the safety intention, the domestic intention and the
locality intention are clear. After buying agricultural products,
we send out questionnaires of Table I. in three grades (yes,
no, unknown).
TABLE I. QUESTIONNAIRE ITEM
Intention Questionnaire item
Q1. Did you interest in deliciousness of the testis?
Gourmet intention
Q2. Did you interest in freshness?
Safety intention Q3. Did you interest in level of safety?
Domestic intention Q4. Did you interest in domestic production?
Locality intention Q5. Did you interest in producing area?
Economy intention Q6. Did you interest in value?
B. Evaluation
1) Discriminating existence or nonexistence of user’s in-
tention by an intention level: Data to acquire topic attention
level are documents of item pages which are browsed by
29Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
37 / 79
17 research participants. The number of commodity page is
503. We classified those pages in the commodity category. In
addition, because the result that research participants bought
food stuff of meat dish are divided into pork and beef, we
classified them as a same commodity category. Sometimes,
information not related to the retrieval intention of the user
has been presented in item page. For example, there are
information on recommended commodity and the commodity
ranking of peculiar to sites. The documents were excluded
since these documents are not relate to user’s intention. The
review which contain on the item page was also excluded.
2) Extraction of topic from the item page: We applied the
method explained in III-B to classified item pages. Among
words which appear frequently, the intention is not considered
from words and symbol of the commodity category which
does not relate to the intention. A bag-of-words was prepared,
excluding these words. LDA is applied to the bag-of-words
using collapsed Gibbs sampling [13]. The scalar value of the
Dirichlet hyperparameter for topic proportions is 0.1. The beta
hyperparameter for each entry of the block relations matrix is
0.1. The number of trials is 30000. Among the 5 patterns (2,
4, 6, 8, 16 topics), 4 topics presented the best performance.
The top 5 words were selected from each of topics.
3) Acquisition of the user’s intention: Research participants
were responded to a questionnaire of Table I. If user answered
YES to the questionnaire of palatability and freshness, the user
is regarded to have gourmet intention.
C. Result
The method explained in paragraph III-B was applied to
extract the topic. Words were labeled based on generality.
1) Extraction of topic attention level: The method ex-
plained in paragraph III-C was applied to extract the topic
attention level. Figure 3 shows the median of extracted topic
attention level. Table II shows the result of the questionnaire
about an intention. ○ is yes. X is no. △ is unknown.
TABLE II. RESULT OF THE QUESTIONNAIRE ABOUT AN INTENTION
User 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Q1. ○ ○ ○ ○ ○ ○ ○ X ○ ○ ○ ○ △ ○ △ ○ X
Q2. ○ ○ X ○ ○ ○ ○ △ X X △ X ○ ○ ○ ○ △
Q3. △ X ○ ○ ○ ○ △ △ ○ X X X X ○ ○ ○ ○
Q4. ○ ○ ○ ○ △ ○ ○ ○ ○ X ○ ○ ○ ○ ○ ○ ○
Q5. ○ ○ X ○ ○ ○ ○ △ ○ X X ○ X X ○ ○ ○
Q6. ○ X ○ X ○ ○ ○ X △ ○ ○ X ○ ○ X X ○
2) Discriminating existence or nonexistence of user’s in-
tention by an intention level: We have verified whether we
can discriminate user’s each intention existence from user’s
intention level by the intention model which is explained in
paragraph III-D. The Table III shows result of the multivariate
logistic regression model. The significance level is less than
5%. The significance level (0.042) of a topic intention relating
a value about a safety intention is extracted from result of the
multivariate logistic regression model. The significance level
(0.042) is a significant value. This result suggests that a safety
intention is connected with topics of price. The significance
level of each intention except a safety intention has been low.
The significant value to estimate user’s intention was not found
by an intention model.
TABLE III. RESULT OF THE MULTIVARIATE LOGISTIC REGRESSION
MODEL
Intention Label Estimate Std.Error z value P r(> |z| )
agricultural chemical
cultivation method
44.99 20068.94 0.002 0.998
brand -494.47 168178.07 -0.003 0.998
domestic domestic, high quality
cook
1074.23 358333.88 0.003 0.998
deliciousness -898.71 305721.58 -0.003 0.998
AIC 12.773
cook 30.8814 21.4463 1.44 0.1499
gourmet
AIC 18.025
agricultural chemical
cultivation method
-7001 741189 -0.009 0.992
economy deliciousness -23338 2467722 -0.009 0.992
locality, value 5922 626767 0.009 0.992
AIC 8
domestic 2.802 1.984 1.412 0.1579
safety value 6.75 3.32 2.033 0.042
AIC 20.537
locality -3.823 2.396 -1.596 0.111
locality deliciousness 8.476 6.372 1.33 0.183
AIC 20.306
V. DISCUSSION
A. Discussion of the intention model
The significant value to estimate user’s intention has not
found by an intention model. In this work, it is assumed user’s
intention is relate to browsing pages of time and contents.
The longer users browse pages, the more they are interest
in the pages. However, the time changes, depending on how
to browse the page. The weighting is considered as one of
the causes which makes the significance probability low. The
proposed method decides weight without considering how
to scan the page. It is investigated that the inspection time
changes depending on how to scan the page. The examination
method focuses on how to browse page using tab or click. The
Table IV shows the counts of how to browse page using tab
or click. If the frequency in which the page is opened with
TABLE IV. THE COUNTS OF HOW TO BROWSE PAGE USING TAB OR CLICK
User 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Click 59 1 0 1 42 18 13 1 18 0 13 21 0 1 16 20 18
Tab 9 19 21 66 1 0 3 39 0 24 0 1 53 36 1 2 0
a new tab is smaller than that with the current tab, we call
those users a “Click type”, otherwise “Tab type”. The click
type are 2, 3, 4, 8, 10, 13 and 14, while the Tab type is 5, 6,
7, 9, 11, 12, 15, 16 and 17. The Table V shows the difference
between the average of the browsing time from the average of
the browsing frequency. It can be said that browsing time and
counts of the tab type are larger than that of the click type.
The difference at the browsing time appears in how to scan the
page. It is necessary to consider the method of the scanning
of the individual when weighting to topics using browsing
time. It can be expected that the higher degree of fit model
is constructed, considering the method of the scanning.
TABLE V. THE AVERAGE OF THE BROWSING TIME AND THE AVERAGE OF
THE BROWSING FREQUENCY
Tab type Click type
Average of browsing time(second) 51.945 42.354
Average of browsing frequency(second) 1.935 1.457
30Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
38 / 79
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
deliciousness
plant varieties
organic farming, soil
deliciousness
locality, price
unknown
unknown
chemical, freshness
locality
foreign, amount
brand
domestic
chemical, organic
method, locality
chemical, inspection
locality, brand
method, chemical
method, chemical
locality
locality
locality, cook
brand
locality, price
reputation
method
brand, cook
locality
domestic
brand
cook
domestic, quality
price
median
label
Figure 3. Median of extracted topic attention level
B. Acquisition of Web browsing history
The proposed method needs access to the web history of
users. It might sound a rather demanding assumption. Users
agreeing to the method are assumed to install an application
to collect Web browsing histories. It is a trade-off between
their advantages and disadvantages. We should consider two
kinds of disadvantages: a burden to install the application, and
a risk to expose their privacy. On the other hand, the users
are released from stresses coming from recommendation of
irrelevant information. We thought the advantages are greater
than the disadvantages.
VI. CONCLUSION
The present paper estimated an intention level to present the
information based on consumer intention on e-commerce sites.
In this method, we extracted item pages which are contained
each commodity using browsing history. We prepared a bag-
of-words from documents which are contained item pages. We
applied LDA to the bag-of-words to extract topic distribution.
We extracted topic attention level by weighting browsing time
and topic distribution. We constructed the intention model us-
ing the multivariate logistic regression model by an important
topic attention level and user’s intention. The user’s intention
is estimated from a topic attention level with the intention
model in an experiment. The significance level is less than
5%. The significance level (0.042) of a topic intention relating
a value about a safety intention was extracted from result
of the multivariate logistic regression model. To improve fit
degree, we consider transition of the pages. In addition, the
questionnaire items do not necessarily correspond to separate
categorize item pages. We plan to review the questionnaire.
REFERENCES
[1] Japan Finance Corporation: Survey of sixth sector industrialization
in agriculture, https://www.jfc.go.jp/n/findings/pdf/topics 111202 1.pdf
[retrieved: March, 2015] (in Japanese).
[2] JC general research: Survey results of the consumption
behavior of vegetables and fruits, http://www.jc-so-
ken.or.jp/pdf/agri/investigate/141203 01.pdf [retrieved: March, 2015]
(in Japanese).
[3] Agriculture, forestry and fisheries. Information Exchange
Network Business: Survey results of awareness and intention
on the production and distribution information of vegetables,
http://www.maff.go.jp/j/finding/mind/pdf/20040820cyosa.pdf
[retrieved: March, 2015] (in Japanese).
[4] Ministry of Agriculture, Forestry and Fisheries: Japanese preference and
taste, http://www.maff.go.jp/j/keikaku/syokubunka/culture/mikaku.html
[retrieved: March, 2015] (in Japanese).
[5] Japan Finance Corporation: Survey of food intention:
https://www.jfc.go.jp/n/findings/pdf/topics 140911a.pdf [retrieved:
March, 2015] (in Japanese).
[6] Yamaguchi. K, and Takahashi. C, Studies on the Preference of Foods :
Part 2 : Relationship between the attribute and the preference of food,
Cookery Science 15.2, 1982, pp. 104-113.
[7] Isojima. A, Analysis of Consumers Needs for Agricultural Products by
Using Text Mining, National Agriculture and Food Research Organiza-
tion Vol.16, No.4, 2010, pp. 38-42.
[8] Isojima. A, Analysis of a Consumer Questionnaire Pertaining to Rice
by using Text Mining. Agricultural Information Research. Vol.15, No.1,
2006, pp. 49.
[9] Isojima. A, Consumer Perceptions and Selection Standards Rice Pur-
chases : An Investigation into Consumer Considerations, Japanese
Journal of Farm Management Vol. 36, No. 3, 1999, pp. 22-32.
[10] C. D. Manning, H. Schutze, “Foundations of statistical natural language
processing.” MIT Press, Cambridge, MA, 1999, pp. 235-260.
[11] D. M. Blei, A. Y. Ng, M. I. Jordan, “Latent dirichlet allocation.” the
Journal of machine Learning research 3, 2003, pp. 993-1022.
[12] Morita. M, and Yoichi. S, “Information filtering based on user behavior
analysis and best match text retrieval.” Proceedings of the 17th annual
international ACM SIGIR conference on Research and development in
information retrieval. Springer-Verlag New York, Inc., 1994, pp. 272-
281.
[13] T. L. Griffiths, M. Steyvers, “Finding scientific topics.” Proceedings
of the National academy of Sciences of the United States of America
101.Suppl 1, 2004, pp. 5228-5235.
31Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
39 / 79
Secure Communication Between OpenFlow Switches and Controllers
Dominik Samociuk
Silesian University of Technology
Institute of Informatics
Email: dominik.samociuk@polsl.pl
Abstract—We study the applicability of different protocols related
to authentication and access control for secure channel between
switches and controllers in Software Defined Networks communi-
cating with OpenFlow. We firstly show possible problems with the
lack of security mechanisms in OpenFlow architecture. Then we
analyze the usability, advantages, drawbacks and implementation
details of Transport Layer Security, Secure Shell and IPSec
protocols as the secure channel medium for OpenFlow commu-
nication between switches and controllers. Finally, we discuss
their possible extensions to authentication and access control
mechanisms.
Keywords–Network security; Secure architecture; OpenFlow,
Software Defined Network.
I. INTRODUCTION
OpenFlow protocol [1] has generated interest in academic
and business society due to the features it offers to architects
and developers of Software Defined Networks (SDN). By
creating a standardized interface to connect switches with
controllers, control-plane logic was moved to a centralized
controller (or controller group). However, even when strictly
adhering to the specification [2], OpenFlow does not enforce
the use of secure communication channel between a switch and
controller. Namely, the entry on the Transport Layer Security
(TLS) usage was introduced in the OpenFlow specification
and then modified. In the latest version, its usage is just
a recommendation (not a ”must-have” requirement). In the
the OpenFlow switch specification ver. 1.4.0 sec. 6.3.3. it
reads: ”The switch and controller may communicate through
a TLS connection”. Moreover, due to evolving nature of the
OpenFlow protocol, many vendors have not fully implemented
this recommendation. Lack of the TLS adoption and problems
with the implementation of the TLS infrastructure leave a
clear path for attackers to infiltrate OpenFlow networks, using
possible attacks described in the following sections. To worsen
the security case, there are no authentication nor access control
mechanisms (except for an ersatz of authentication in TLS).
Without forcing a secure communication channel, Open-
Flow risks repeating the mistakes of other management pro-
tocols, designed basing on the assumptions that the link
and infrastructure are secure (e.eg. Telnet, SNMPv2, TFTP).
Of course, in production environment they must have been
replaced with their safe versions (SSH, SNMPv3, SFTP, re-
spectively). One of the proposed solutions for OpenFlow is
controlling switches through the Internet in the architecture
of branch-office network or offering switch management as
”security-as-a-service”. If the secure communication channel,
authentication and access control are not enforced, OpenFlow
will not be able to develop into the above-described roles and
will be replaced by a secure protocol (like in the mentioned
transition from Telnet to SSH) or by another, secured version
of the protocol (like in transition from SNMPv2 to SNMPv3).
In this paper, we compare the possibility of implementing
different authentication and access control mechanisms over
a secure channel in OpenFlow communication with three
popular protocols, namely Transport Layer Security, Secure
Shell (SSH) and IPSec.
Transport Layer Security [3] and its predecessor, Secure
Sockets Layer (SSL), are cryptographic network protocols for
securing data communication. TLS provides confidentiality
and integrity of data, as well as the authentication of the
server, and sometimes of the client. It is based on asymmetric
encryption and can be deployed in two modes, namely as:
• X.509 certificates and Public Key Infrastructure cryp-
tosystem to provide authentication, encryption, in-
tegrity and non-repudiation using public and private
key cryptography and digital certificates;
• Web of Trust architecture – decentralized authentica-
tion method, in which there is no hierarchical structure
of the authenticating organizations and trust of each
certificate is the sum of the signatures by the other
members of the web, signed under this certificate.
SSH [4] is a common name for the whole family of
protocols, not just terminal. It is also used for file transfer (SCP,
SFTP), remote control of resources, tunneling and other appli-
cations. A common feature of all these protocols is identical
to the SSH data encryption technology and user recognition.
It is possible to configure SSH tunnels to transfer unencrypted
traffic on the network through an encrypted channel.
IPSec [5] is a set of protocols for implementing secure
connection and encryption exchange of keys between hosts.
IPSec can be used for protecting the transmission in three
modes:
• host-to-host – between pair of hosts;
• network-to-network – between pair of the security
gateways;
• network-to-host – between the gateway and a host.
IPSec consists of at least two channels of communication
between connected devices: (a) the exchange channel, through
which data associated with authentication and encryption
(keys) is transmitted and (b) the channel (one or more) that
carries packets transmitted over the already secured line.
This paper is organized as follows. In Section 2, we
discuss related works. Section 3 describes the secure channel
and authentication problems with OpenFlow in the recent
specification. It also presents some possible attacks and their
32Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
40 / 79
effects. In Section 4, we present technical details about the
proposed solutions, including Transport Layer Security with
PKI architecture, Secure Shell tunneling and IPSec protocol.
In Section 5, a brief security analysis of the security of the
proposed solutions is carried out. Section 6 concludes the paper
and presents our planned future work on the subject, including
the implementations of new solutions and experiments on PL-
LAB2020 - a large testbed and experimental network being
built in Poland right now.
II. RELATED WORK
So far, the following works have been focused on the secure
SDN architecture using OpenFlow.
FlowVisor [6] acts as a transparent proxy between con-
trollers and switches in order to apply limitations to the rules
created by controllers. It creates slices (combinations of switch
ports, MAC addresses, IP addresses, port addresses or ICMP
type) of network resources and delegates the control of slices
to different controllers. The role of FlowVisor is to isolate the
effect of rewritten rules to the specific slice of the network i.e.,
one slice cannot control another’s traffic.
A similar concept is FortNOX [7] – a software extension
developed on NOX controller to check the flow rule contradic-
tions in real time. It uses the role-based authorization on Open-
Flow applications (in this case, something that wants to modify
the network traffic using the OpenFlow protocol, e.g., fire-
walls, intrusion prevention systems). The difference between
FortNOX and FlowVisor is that FortNOX is a single controller
software, that executes parallel applications, while FlowVisor
runs apart from controllers (usually on a different host). Both
of these solutions restrict intrusted controllers/applications
from introducing security threats. However, they rely on the
assumption that the OpenFlow protocol and its communication
channel are secure.
Another approach is to provide the security to Software
Defined Networks using OpenFlow. NICE [8] Distributed
Denial of Service protection infrastructure-as-a-service is a
distributed vulnerability detection tool based on attack graph-
based analytical models and reconfigurable countermeasures.
Moving Target Defense [9] in an OpenFlow environment is
a mechanism to change internal hosts’ IP addresses frequently
and to mitigate attacks and reconnaissance from external
network.
All the mentioned works are meant to provide the security
to an OpenFlow-based SDN architecture, but their underlying
assumption is that there are no network design vulnerabilities
with the protocol. A comprehensive OpenFlow vulnerability
assessment was presented in [10]. The list of vulnerabilities
contain the lack of TLS adoption, flow enforcement, denial
of service risk and controller vulnerabilities. The first three
of them can be fully mitigated (or significantly weakened at
least) by using authentication and access control mechanisms,
which are examined in this paper.
III. OPENFLOW SECURE CHANNEL PROBLEMS
The OpenFlow specification in version 1.0 contains the
requirement about the use of TLS [11]. However, the next
version changed this requirement from ’must’ to ’should’.
This is also the case of the current version (1.4). There
is a noticeable lack of support for TLS in current SDN
switches and controllers. Table I shows TLS support offered
by OpenFlow equipment vendors [10][12].
TABLE I. TLS SUPPORT IN OPENFLOW BY VENDORS.
VENDOR TLS Support
HP switch No
Brocade switch Controller port only
Dell switch No
NEC switch Partial
Indigo switch No
Pica8 switch Only new versions
Open vSwitch Yes
NOX controller No
Brocade Vyatta controller Yes
POX controller No
Beacon controller No
Floodlight controller No
MuL controller No
FlowVisor No
Big Network controller Yes
Open Source controllers (f.e. Ryu, OpenDaylight) Yes
The usage of Transport Layer Security has also its impact
on the preparation and maintenance of Software Defined
Networks based on the OpenFlow protocol. In particular, this
includes generating controller and switch certificates, signing
certificates with private keys, installing correct keys and certifi-
cates on devices. Assuming a topology distributed in different
locations, though not connected to the Internet, technicians
must prepare all components of Public Key Infrastructure and
provide its security.
While the lack of TLS support is feasible in secured
networks (such as data centers) where the access to physical
devices is difficult, it becomes a serious security vulnerability
in architectures similar to campus-style or branch-offices de-
ployments, in which access to the network is less restricted.
In the role of the management protocol in the ”security-as-a-
service”, it can be the case when OpenFlow is transmitted by
an untrusted ISP (e.g., the client is in the country interested
in intercepting transmission). In such scenarios we cannot
neglect the possibility of an attacker placing a device on
communication path between the switch and the controller, or
simply copy the flow to his/her machine (Figure 1). He/she
is able then to get the configuration, insert or delete rules
to modify/record sensitive data flow (configured using the
OpenFlow device). Additionally, an attack can be performed
without any observable differences from normal transmission,
i.e., the attacker acts as a transparent proxy (compare this
with the FlowVisor proxy modification of the rules through
the process of forwarding messages). This type of attacks
(so called ’man-in-the-middle’ attacks) were very popular and
successful before, [13][14]. The Software Defined Networking
may decrease the difficulty of a full exploitation and may allow
the attacker to automate the process. In the SDN, the man-
in-the-middle attacks are arguably worse than in non-software
defined networks, due to the lack of necessity to the sniff traffic
to obtain plain-text credentials or the possibility to reconfigure
all groups of devices in a single attack.
It is important to note that assuring a secure communication
between the switch and controller is not enough – we should
also authenticate all devices connected to the controller or
switch. For instance, for the following two reasons the au-
thentication and access control may be needed:
• to limit the possibility of adding a bogus device (e.g.,
33Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
41 / 79
Figure 1. Man-in-the-middle attack scenario.
adding a new switch or controller to the OpenFlow
domain);
• to group devices in federation (compare e.g., the
Ofelia project, [15]), when different groups should be
configured with different rules.
With full TLS implementation and authentication mech-
anism we can assure safety of the hosts and messages in
transmission. We are unable, however, to detect switches that
operate erroneously on rules in databases. The only solution
to maintain the same view of traffic that flows through the
network on the controller and switch seems to be dumping
regularly and inspecting the flow tables on all hosts in the
OpenFlow architecture. A potential solution described in [10]
is based on generating keep-alive messages with checksums of
flow tables that switches sent to controller.
IV. SECURE CHANNEL CREATION AND AUTHENTICATION
In this section, we describe three possibilities of creating
the secure channel for communication between switches and
controllers using OpenFlow. We also investigate how to fur-
ther secure the OpenFlow architecture with an authentication
mechanism that reduces the possibility of spoofing a device
with a rogue switch or controller instead of any mitigation of
eavesdropping.
A. Transport Layer Security
Deploying Transport Layer Security as an authentication
and access protocol in the OpenFlow-based architectures was
done by vendors using the public key infrastructure (PKI).
The Web of trust architecture was neglected due to the small
amount of devices in the authentication domain which could
result in the vulnerabilities described in [16].
The proposed solution is an example of the peer encryption.
The number of exchanges of the keys in such systems is
proportional to the square of the number of users of the system
(precisely, proportional to
n(n−1)
2
, where n is the number of
users of the system). As already discussed, the reliability of the
key distribution is essential for the credibility of the system.
The solution to the number of the needed exchanges of keys
and the need to ensure their authenticity was an application of
the principle of the implied trust.
In cryptographic systems, there might be available some
certification institutions, to which we have trust. This trust is
supported by, for example, their protection level, their regular
auditing, etc. If all users of the system have trust in the
certification center, it is assumed that they also have trust in
each other. In practice, this means that a user should know
the address of the certification center and have its public key.
In this way the user can reliably verify the authenticity of the
received documents or exchange encrypted information with
a reliable key from the communication partners, published by
the certification center. In the PKI infrastructure, the trusted
certification center is known as the Certificate Authority (CA).
To provide an opportunity to exchange keys between
different systems, the cryptographic standard X.509 has been
introduced also for certification. A certificate contains not only
the owner of the public key signed by a certification authority,
but also the information about the owner and other fields
predicted by the standard X.509. Not only must CA be able to
issue the certificate, but also to public and cancel the certificate,
for example due to a theft or other security breach.
Equally important, as the implementation of the infrastruc-
ture, is developing and implementing the procedures for han-
dling of the certificates. These procedures are called Certificate
Practice Statement and include, among others:
• rules for issuing, verification and control the distribu-
tion of certificates;
• rules for cancellation of certificates;
• ability to recover private keys;
• methods for securing the infrastructure.
Especially in the case of PKI systems, the procedures are pillar
of security.
In order to implement a TLS based secure channel and
authentication mechanism, the administrator must build all the
infrastructure to support it, i.e. the PKI (Figure 2). It will
consist of elements such as:
• Each user of the system (in the case of an OpenFlow
architecture, each switch and controller) posses key
pair of private and public, stored in a secure space.
• Users of the system adopt a standard for certificate to
the exchange keys.
• The existence of the certifying authority (internal or
external, when developing OpenFlow in security-as-
a-service manner) providing the issuance, invalidation
and publication of certificates.
• Implementation of procedures to ensure the safety of
the system.
Figure 2. Architecture of the secured channel based on TLS protocol.
A simple connection between the switch and the controller
illustrating the handshake with authentication by means of the
Transport Layer Security mechanism consists of the following
steps (by C we denote the controller; by S – the OpenFlow
switch):
• S to C: sends ClientHello
34Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
42 / 79
The switch sends to the controller a message contain-
ing, inter alia, supported version of TLS protocol, sup-
ported methods of data encryption and compression,
and session ID. This message also contains a random
number which later will be used for key generation.
• C to S: sends ServerHello
The controller responds with a similar message, in
which it returns to the switch the selected parameters
of the connection: TLS protocol version, supported
types of encryption and compression, and a random
number.
• C to S: sends Certificate
The controller sends its certificate allowing the switch
to verify its identity.
• C to S: sends ServerKeyExchange
The controller sends its public key. The type and
length of the key is determined by the type of al-
gorithm in the previously sent message.
• C to S: sends ServerHelloDone
The controller notifies that the switch can move to the
next phase of the secure channel creation setup.
• S to C: sends ClientKeyExchange
The switch sends to the controller an initial session
key encrypted with the public key of the controller.
Using the previous messages, the two random numbers
(one for switch and another for controller) as well
as pre-determined by the switch session key, both
sides generate a session key used for the actual data
exchange. The key is generated using a symmetric
algorithm (typically DES). However, it is set in a safe
way and known only for communicating parties.
• S to C: sends ChangeCipherSpec
The switch informs the controller that it can switch to
encrypted communication.
• S to C: sends Finished
The switch sends this messages to report readiness to
receive encrypted messages.
• C to S: sends ChangeCipherSpec
The controller notifies that it obeyed the request -
from now on the controller will only send encrypted
information.
• C to S: sends Finished
The message is sent over the secure channel to check
reliability of the used mechanism.
As shown in the steps in the previous section, the default
TLS mechanism provides only server authentication, resulting
in authenticating the OpenFlow controller and leaving the
possibility of spoofing the OpenFlow switches (grabbing con-
figuration or modifying rules and sending them not so secured
switch if possible). However, there are methods to authenticate
the client switch. For this purpose, three additional messages
can be used:
• C to S: sends CertificateRequest
After submitting controller’s certificate server notifies
the client that it would like to receive a certificate from
the OpenFlow switch
• S to C: sends Certificate
After receiving the message ServerHelloDone the
switch sends its certificate
• S to C: sends CertificateVerify
The switch must confirm that it actually has the private
key corresponding to the transmitted certificate. To
prove this, the switch signs with its private key digest
of all previously established connection parameters
and sends it using this message.
B. Secure Shell
Another approach to encryption and authentication can be
using the automatically generated keys and trust-of-first-use
method as in the Secure Shell protocol. All switches are treated
as SSH clients and the controller is treated as their server.
The client connects to the server, authenticates with the key.
In the process, it authenticates also the server key. Then the
secure tunnel is created between OpenFlow’s communication
port on the server side (the controller) and configured port on
switch. The switch sends unencrypted traffic to a local port (it
is assumed that insecure architecture starts when a packet is
leaving the network card and the device itself is trusted), and
then the traffic is transmitted to the right port on the controller
using the encrypted and secure tunnel (Figure 3). In this way,
we limit the possibility of eavesdropping messages directly
to the locally bugging switch and controller. We also use the
authentication against the possibility of spoofing devices.
Figure 3. Secure SSH tunnel transporting the OpenFlow data.
We propose two similar authentication-related solutions,
but with different level of security and configuration required.
The first one is the usage of automatically generated keys
and automatic acceptation of the connection by a client before
the typical SSH authentication process. This will reduce the
possibility of performing and attack on the transmission to
the small window of the first communication between the
switch and controller, but it does not require any additional
configuration. Alternatively, the more secure (but requiring
the involvement of the administrator) solution would be to
verify the public key thumbprint of each device when con-
necting to the server for the first time. Additional overhead
of the administrative work when adding a new switch to
the OpenFlow topology will completely reduce the possibility
of eavesdropping transmission of the OpenFlow messages
tunneled via SSH.
C. IPSec
The last analyzed option of authentication and creation of
a secure channel between the switch and controller over the
OpenFlow communication is the IPSec protocol, in particular
the host-to-host architecture. IPSec in the host-to-host configu-
ration (which connects two hosts without the need of additional
devices, see Figure 4), creates the secure channel between them
and allows to authenticate each other, while the only needs are
connections dedicated to the other side. The process can be
summarized in the following five steps:
35Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
43 / 79
1) Configuring switches and controllers.
In this phase the network administrator must prepare
the information (IP addresses of both hosts, encryp-
tion key generator, pre-shared key used to initiate
the connection and exchange generated keys during
secure transmission) and then use them during config-
uration of each device. This configuration generally
will be prepared once and then copied between hosts.
2) IKE phase one.
The purpose of this phase is to authenticate hosts
and set up the first secure channel for IKE exchanges.
Following functions are performed: the authentication
and protection of the hosts identities, the negotiation
of the IKE policy to protect the exchange, the usage
of Diffie-Hellman authenticated exchange to obtain
matching shared keys, the setting up of the first
secure channel for IKE phase two. Phase one has two
operational modes: main and aggressive. The main
mode consists of three exchanges (agreement of the
algorithm and hashes used to secure communication,
generation of secret keys components used to pass-
ing random numbers - nonces to prove identities,
last exchange is the verification of the other side’s
identity). The aggressive mode has fewer exchanges
(we obtain second IKE phase quicker) but some of
the information must be exchanged before there is a
secure channel available.
Due to the security aspect of the work, we prefer the
main mode in the planned implementations.
3) IKE phase two.
In this step the IPSec tunnel is created. The fol-
lowing functions are performed: the negotiation of
protected parameters, establishing of security asso-
ciations, renegotiation IPSec SA to ensure security
(additional authenticated Diffie-Hellman exchanges
can be performed).
4) Data transfer.
After the IKE phase two, the IPSec tunnel is cre-
ated and OpenFlow packets can be encrypted and
decrypted using the encryption mechanism specified
in the configuration, resulting in authenticated secure
channel transmission.
5) IPSec tunnel termination.
After the successful transmission, the tunnel can be
terminated by deletion or by timing-out.
Figure 4. Secure IPSec tunnel transporting the OpenFlow data over an
encrypted channel.
While TLS and SSH operate in the application layer,
IPSec is a scheme operating in the network layer. Hence,
only IPSec protocol protects any application traffic over an
IP network. Therefore, an implementation for the OpenFlow
protocol can be relatively easily migrated to another protocol
or (due to evolving nature of OpenFlow) next specifications
requirements.
V. SECURITY ANALYSIS
We analyzed benefits and drawbacks in the area of security
of the proposed solutions. TLS, SSH and IPSec reduce or
prevent the possibility of performing the following attacks:
• sniffing (lack of confidentiality),
• data modification,
• identity spoofing, password-based and application-
layer attacks,
• man-in-the-middle attacks,
• denial-of-service attacks.
However, there are a few concerns regarding each of the
solutions.
Transport Layer Security relies on secrecy of private keys
and Certificate Authority trust, so assuring the security of these
parts is the mission-critical aspect of maintaining this type
of architecture. Some significant attacks against TLS include
FREAK [17], BEAST [18] and CRIME and BREACH [19]
attacks. However, assuming a proper implementation of TLS
and its newest version, it is regarded as safe. Theoretically
TLS can be compromised using SSL-Striping and SSL-Spitting
attack, [20], but due to the newness of software-defined net-
working with TLS-secured OpenFlow, the attacks has not been
confirmed and sufficiently researched yet.
As mentioned above, Secure Shell can be deployed using
public and private keys or pre-shared key. Regardless of the
option used, secrecy of keys (and CA if used) is the crucial
security aspect. As regards possible attacks, using the revised
version SSH-2 is assumed to be secure, however, theoretical
vulnerability was discovered in [21] for the default encryption
mode CBC. Therefore, we recommend the usage of CTR mode
in the implementation. Another security concerns are that some
institutions are able to decrypt the SSH traffic (see [22]). The
details associated with such attacks were not released.
IPSec, similar to TLS and SSH relies on the secrecy of pre-
shared key. However, the critical aspects are the randomness
of the encryption key generator and security of the encryption
and hash algorithm. As for this work, from available in IPSec
implementation algoritms we assume sha1 as hash algorithm
and 3DES, AES as encryption algorithms are secure. There
are known attacks on IPSec when other than recommended
solutions were applied [23]. Similar to SSH, there are alle-
gations that some institutions have been working actively to
insert vulnerabilities into IPSec implementations, [24].
VI. CONCLUSION AND FUTURE WORK
In this paper, some possible solutions to the lack of au-
thentication, access control and creation of the secure channel
over the OpenFlow protocol were investigated. As argued,
implementing TLS (as in the OpenFlow specification rec-
ommendations) does not address configuration problems for
network operators. With the core idea of increased security
of the OpenFlow transmission, a novel utilization of known
Internet security systems was proposed.
By comparing TLS, SSH and IPSec, it was demonstrated
that, in relation to the OpenFlow architecture usage, each
of the proposed protocols has its own strengths (i.e. the
ease of implementation in IPSec, or conformation with the
specification in TLS) and weaknesses (possible attacks). The
36Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
44 / 79
paper showed that implementing any of the proposed solutions
will result in increased security and reduction or prevention
against numerous attacks on the OpenFlow protocol.
As for the future work, we are planning to utilize the PL-
LAB2020 laboratory, [25], to implement each solution and
analyze the security concerns mentioned in the article while
verifying the performance of the protocols. The PL-LAB2020
laboratory, which is under construction now, will consist of six
geographically dispersed nodes associated with leading Polish
research and academic centers:
• National Institute of Telecommunication (NIT),
• Warsaw University of Technology (WUT),
• Poznan Supercomputing and Networking Center
(PSNC),
• Silesian University of Technology (SUT),
• Gdansk University of Technology (GUT),
• Wroclaw University of Technology (WrUT),
connected via dedicated 2*10Gb/s fiber links. The nodes
will be equipped with specialized devices for carrying out
research in several directions, including the Software Defined
Networking. In particular, there will be over 30 OpenFlow
switches from at least 3 different vendors and a few OpenFlow
controllers creating different technology domains distributed
over 5 locations of PL-LAB2020 infrastructure (Figure 5). To
validate the performance and analyze security of the studied
solutions, PL-LAB2020 will also consist of several servers
with Data Plane Development Kit and three network traffic
generators and analyzers with 10Gb/s interfaces, placed in
different locations.
Figure 5. Architecture of PL-LAB2020.
ACKNOWLEDGMENT
This work has been supported by the National Centre
for Research and Development under the European Regional
Development Fund, Grant No. POIG.02.03.01-00-104/13, PL-
LAB2020 project.
REFERENCES
[1] N. McKeown and et al., “Openflow: enabling innovation in campus net-
works,” ACM SIGCOMM Computer Communication Review, vol. 38,
no. 2, 2008, pp. 69–74.
[2] “Openflow switch specification: Version 1.4.0,” URL:
https://www.opennetworking.org/images/stories/downloads/specification/
openflow-spec-v1.4.0.pdf [accessed: 2015-02-28].
[3] T. Dierks, “The transport layer security (tls) protocol version 1.2,” 2008,
pp. 4–68.
[4] T. Ylonen and C. Lonvick, “The secure shell (ssh) protocol architec-
ture,” 2006, pp. 4–26.
[5] N. Doraswamy and D. Harkins, IPSec: the new security standard for
the Internet, intranets, and virtual private networks. Prentice Hall
Professional, 2003.
[6] R. Sherwood and et al., “Flowvisor: A network virtualization layer,”
OpenFlow Switch Consortium, Tech. Rep, 2009, pp. 1–13.
[7] P. Porras, S. Shin, V. Yegneswaran, M. Fong, M. Tyson, and G. Gu, “A
security enforcement kernel for openflow networks,” in Proceedings of
the first workshop on Hot topics in software defined networks. New
York: ACM, 2012, pp. 121–126.
[8] C.-J. Chung, P. Khatkar, T. Xing, J. Lee, and D. Huang, “Nice: Network
intrusion detection and countermeasure selection in virtual network
systems,” IEEE transactions on dependable and secure computing, no. 4,
2013, pp. 198–211.
[9] J. H. Jafarian, E. Al-Shaer, and Q. Duan, “Openflow random host
mutation: transparent moving target defense using software defined
networking,” in Proceedings of the first workshop on Hot topics in
software defined networks. New York: ACM, 2012, pp. 127–132.
[10] K. Benton, L. J. Camp, and C. Small, “Openflow vulnerability assess-
ment,” in Proceedings of the second ACM SIGCOMM workshop on
Hot topics in software defined networking. New York: ACM, 2013,
pp. 151–152.
[11] “Openflow switch specification: Version 1.0.0,” URL:
https://www.opennetworking.org/images/stories/downloads/sdn-
resources/onf-specifications/openflow/openflow-spec-v1.0.0.pdf
[accessed: 2015-02-28].
[12] Y. Patil, “Vulnerability analysis of openflow control channel,” 2014, pp.
1–2.
[13] F. Callegati, W. Cerroni, and M. Ramilli, “Man-in-the-middle attack to
the https protocol,” IEEE Security and Privacy, vol. 7, no. 1, 2009, pp.
78–81.
[14] K. Ouafi, R. Overbeck, and S. Vaudenay, “On the security of hb# against
a man-in-the-middle attack,” in Advances in Cryptology-ASIACRYPT
2008. Springer, 2008, pp. 108–124.
[15] “OFELIA PROJECT homepage,” URL: http://www.fp7-ofelia.eu/ [ac-
cessed: 2015-02-28].
[16] N. Ferguson and B. Schneier, Practical cryptography. Wiley New York,
2003, vol. 141.
[17] B. Beurdouche and et al., “A messy state of the union: Taming the
composite state machines of tls,” in IEEE Symposium on Security and
Privacy. IEEE, San Jose, 2015, pp. 1–16.
[18] T. Duong and J. Rizzo, “Here come the ninjas,” Unpublished
manuscript, 2011, p. 4.
[19] D. Goodin, “Crack in internets foundation of trust allows https session
hijacking,” Ars Technica, 2012, pp. 1–2.
[20] M. Marlinspike, “New tricks for defeating ssl in practice,” BlackHat
DC, February, 2009, pp. 1–114.
[21] “SSH Vuln Note,” URL: http://www.kb.cert.org/vuls/id/958563 [ac-
cessed: 2015-02-28].
[22] “Prying Eyes: Inside the NSA’s War on Internet Security,” URL:
http://www.spiegel.de/international/germany/inside-the-nsa-s-war-on-
internet-security-a-1010361.html [accessed: 2015-02-28].
[23] J. P. Degabriele and K. G. Paterson, “Attacking the ipsec standards in
encryption-only configurations.” in IEEE Symposium on Security and
Privacy, vol. 161, Oakland, 2007, pp. 335–349.
[24] “Secret Documents Reveal N.S.A. Campaign Against Encryption,”
URL: http://www.nytimes.com/interactive/2013/09/05/us/documents-
reveal-nsa-campaign-against-encryption.html [accessed: 2015-02-28].
[25] URL: http://http://www.pllab.pl/ [accessed: 2015-02-28].
37Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
45 / 79
Cloud Assisted Live Video Streaming over DHT Overlay Network
Pheng Un Lim and Hwangkyu Choi
Department of Computer Science and Engineering
Kangwon National University
Chuncheon, Korea
e-mail: {limphengun, hkchoi}@kangwon.ac.kr
Abstract—Many researches have been proposed to solve the
scalability, the availability and the low-latency problems of
peer-to-peer live video streaming; however, the problems are
still not yet completely solved. In this paper, we propose a
cloud assisted live video streaming over the distributed hash
table (DHT) overlay network to tackle these problems. Our
system is based on a scalable DHT network, which structures
into a mesh-based overlay, to efficiently share the video stream.
Moreover, cloud assist is introduced in order to maintain the
availability, the low-latency and the scalability of the system.
In this work, cloud server is also part of the DHT network. The
role of the cloud server is to assist other nodes when the
number of nodes which received video segment is less than the
specific threshold. The cloud servers directly receive video
segments from video server so the quality and the availability
of the video segments can be guaranteed. The evaluation
results show that with the help of cloud assist, the availability
and the low-latency of the video segments are greatly improved
if cloud has enough bandwidth.
Keywords-video live streaming; cloud computing; peer-to-
peer; DHT overlay network.
I. INTRODUCTION
Because of scalability, availability and low-latency
problems, the current live video streaming still cannot meet
the user demand. Many studies have been tried to solve
these problems. One of the most recent researches which
tried to solve these problems is peer-to-peer (P2P) live video
streaming over the DHT network. Yet, the availability and
the low-latency problems are still the main obstacles of this
system because it fully depends on the user device’s
capacity. In P2P live video streaming if the user device’s
capacity is poor, other clients will not get the desirable
quality of the video stream.
Peer-to-peer systems are distributed systems in which
each node has equivalent functionality and without any
centralized control or hierarchical organization [1]. The core
operation of P2P system is the efficiency of locating each
node in the network. The distributed hash table (DHT) is
one of the most efficient methods which can be used to
locate the node in mesh-based P2P network. DHT is a
decentralized system that provides a lookup service similar
to a hash table. The DHT node uses key to lookup for the
value which associated with it. The advantages of DHT are:
high scalability, reliability, and self-organizing [2].
However, only few works have been focused on using DHT
with mesh-based mechanism. Thus, cloud assisted live
video streaming over DHT overlay network is proposed in
this paper.
In this system, the P2P network is based on a DHT using
the available protocols such as Chord [1], Pastry [3],
Kademlia [4], etc. A cloud server is considered as a node,
the same as other users’ devices. The only different is that
cloud server has a big and stable capacity. The role of the
cloud server is to assist other nodes when the numbers of
available segments are less than the specific threshold.
The rest of this paper is organized as follow: Section 2
talks about the related works. The system architecture of the
new proposed system, buffer map management, data push
and pull among peers, and cloud assist scenario are
discussed in Section 3. The system performance evaluation
is discussed in Section 4. Finally, the conclusion is
presented Section 5.
II. R
ELATED WORKS
There are many existing researches have been studying
on P2P video live streaming; however, most of those works
focused on tree-based, mesh-based with gossip protocol and
others, and hybrid structure such as [5]-[7]. Only few
researches focused on using mesh-based with DHT [2] and
cloud assist in P2P live video streaming [8]-[10].
[2] is one of the most recent research which built a P2P
live video streaming over DHT network and using chord
protocol. This work was proposed to solve the problems of
scalability, availability, and low-latency in P2P live video
streaming systems. The performance of this system is far
better than that of mesh-based and tree-based system.
However, it fully depends on user devices’ capacity which
is small and not stable (nodes join and leave the system any
time) and also the video server can become a bottleneck
when the number of users is suddenly increased. In order to
maintain the video quality and the scalability of the system,
and also to deal with the inconsistency of the clients’
devices, cloud assist has been proposed in our paper.
[8] proposed a cloud-based service called “AngelCast”.
AngleCast is proposed to solve the limitation of the
delivered video quality. The client nodes in AngleCast are
arranged in tree-based structure. [9] proposed Cloud-
38Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
46 / 79
Assisted P2P Live Streaming (Clive) to tackle the
bottlenecks in the available upload bandwidth at media
source and inside the overlay network that can limit the
quality of the server (QoS). Cloud-Assisted Live Media
Streaming (CALMS) [10] books and adjusts adaptively
cloud server resources in a fine granularity to accommodate
temporal and spatial dynamics of demands from live
streaming users.
As most of the proposed cloud assisted P2P live video
streaming researches tend to solve the problems of tree-
based, mesh-based or hybrid structure, only few works are
focused on hybrid structure with DHT network. Thus, in
this paper, a cloud assisted live video streaming system over
DHT overlay network is proposed.
III. S
YSTEM ARCHITECTURE
The cloud assisted video live streaming system is built
over DHT overlay network, as shown in Figure 1. In this
system, cloud servers join the DHT network as a node the
same as other clients and receive video segments directly
from the video server.
Figure 1. Cloud assisted live video streaming over DHT overlay network
A. Buffer Map Management
In P2P live video steaming system, the video stream is
divided into segment or chunk which identified by the
sequence numbers, name as segment ID [11]. Each node has
a buffer to catch the video segment. A buffer map, which
stores several video segments, as shown in Figure 2, is an
abstract description of this buffer. Buffer map consists of a
sequence of {0, 1} to indicate the buffered and empty states
of the segment [12]. While playing the video stream, node
can refer to its buffer map in order to know which segment
is missing and which segment is already buffered. Thus,
node can request the missing segment in advance. In this
system, each node stores the buffer map information list of
the segment which hash key’s ID equals to or immediately
succeeds the node’s ID.
Figure 2. Live streaming scenario
One video segment can have more than one hash key
because segment belongs to node in different location will
generate different hash key. Hash key contains channel
information, program information, segment number and
location information, as shown in Figure 3.
Figure 3. Hash key
The buffer map information consists of the IP address,
starting segment and the current buffer map, as shown in
Figure 4. The buffer map information list stores the buffer
map information in the descending order of the starting
segment number.
Figure 4. Storing buffer map info process
When node receives segment from server or from other
node, this node has to register its buffer map information to
the DHT network. The processes of registering its buffer
map information are: first, hash key for this segment is
created. This hash key is used to find the node to store the
buffer map information. The same segment, which has the
same hash key, will be stored in the same node. Then, the
DHT method, put (key, buffer map info), is used to find the
appropriate node to store the buffer map information, as
shown in Figure 4.
B. Data Delivery Management
During playing the video stream, each node checks their
buffer map information in order to know whether there is a
39Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
47 / 79
missing segment. If there is a missing segment, node uses
the missing segment’s ID, channel information, program
information, and node’s location information to create a
hash key. Then, this hash key is used to find the node which
stores the buffer map information list of this missing
segment. After that, the requested node requests the buffer
map information list from that node. The DHT method, get
(key), is used to retrieve the buffer map information from
specific node, as shown in Figure 5.
Figure 5. Request segment among peers
Finally, the requested node uses the received buffer map
information list to find the node which has the missing
segment and sends the request to that node.
C. Cloud Server Assist Scenario
The role of cloud server is to assist other nodes in the
DHT network. Cloud server will assist other nodes in case
the number of nodes which has the video segment is less
than the threshold. The threshold is referred to the number
of nodes which receive segment directly from server that
can effectively share the video segment to other nodes in the
network.
Figure 6. Cloud assist scenario
As illustrates in Figure 6, for each segment, cloud server
requests the buffer map information list from the node
which stores this list. Then, cloud server uses the buffer map
information list to count the available segments. If the
numbers of available segments is less than the threshold,
cloud server generates hash key and sends its buffer map
information to the node which stores the buffer map
information list of this video segment. Thus, other nodes
can request the video segment from cloud server. After
receiving the segment, those nodes also register their buffer
map to the network to share the segment to other nodes.
Therefore, the cloud server will not be overloaded by many
requests.
IV. P
ERFORMANCE EVALUATION
In order to evaluate the system performances, a
simulation program has been developed. In our experiment,
we set the upload bandwidth of the server and cloud to
40,000kb/s and upload and download bandwidth of nodes to
600kb/s. The video size is set to 120 and segment size is
300kb, that one segment can be played for 1 second. Thus,
the video length is 2 minutes. The threshold is set to 16
percent of the total nodes. The buffer size is set to 10. After
the upload bandwidth of the clouds and the nodes which
have the video segment are full, the requested nodes are
queued and will request the video segment again after the
bandwidth is available. The number of nodes join and leave
the network are set to the exponential distribution with mean
value of 2 percent and 1 percent of total nodes every half
second, respectively. The delay time of each segment is set
to 3 seconds, means that if all nodes receive the segment
during 3 seconds, the user can watch the video stream
smoothly. We run each experiment five times and report the
average as the final experimental results.
A. Latency
As illustrates in Figure 7, the average delay time of the
segments without the help of cloud assist exceed 3 seconds
if compare to when using cloud assist, which only exceed 3
seconds, when the number of nodes reach 3000.
Figure 7. Delay time vs. number of nodes
As mention in the experiment setup, the threshold is set
to 16 percent of total nodes. Thus, even when the number of
nodes reach 8000, the video server still have enough
bandwidth to send the video segment to threshold nodes (the
total bandwidth which is required for sending the video
segment to the threshold nodes is only 37,500KB). However,
because nodes dynamically leave and join the network, the
delay time of each segment cannot be maintained. For
cloud-assist method, the average delay time of each segment
exceed 3 seconds when number of nodes reach 3,000
because cloud server does not have enough bandwidth.
40Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
48 / 79
Figure 8. Cloud bandwidth vs. number of nodes
Figure 8 shows the amount of cloud bandwidth which is
needed for maintaining the delay time of the network, i.e., if
this amount of cloud bandwidth is used for each method, the
delay time will not exceed 3 seconds.
B. Availability
Figure 9 shows the number of remained nodes when the
number of nodes in the network is increased. The remained
nodes refer to the nodes which do not receive the video
segment after the delay time (the delay time is set to 3
seconds in this experiment) is passed. The number of
remained nodes of no-cloud method is worse than cloud-
assist method.
Figure 9. Number of remained nodes vs. number of nodes
Even the delay time, as shown in figure 7, of both
methods has a little different, the number of remained
methods of no-cloud assist is far bigger than that of cloud-
assist methods. For instance, when the number of nodes
reaches 3000, the delay time of no-cloud method is 3.49 and
the delay time of cloud-assist method is around 3.05.
However, according to figure 9, for no-cloud assist, after the
delay time 3 (after 3 second) passed, the number of
remained nodes is 413 compare to the remained nodes of
cloud-assist methods is only around 5.
V. C
ONCLUSION AND FUTURE WORK
In this paper, cloud assisted live video streaming over
DHT overlay network has been proposed. The DHT
network can be used to solve the problem of scalability.
Moreover, because cloud server is consistent, has big
capacity, and receives video segments directly from video
server, the availability of the video segments and the low-
latency can be enhanced. The performance evaluation
results show that with the present of cloud assist, the
performances of the system are greatly improved in both
availability and low-latency. We only build a simulation
program to compare the performance of the system when
there is no cloud assist and when there is a cloud assist. In
the future, a full function DHT simulation process should be
conducted. In addition, the threshold value and the amount
of cloud capacity which is required for assisting the failure
nodes also need to be studied in details in order to maximize
the system performance and minimize the cost.
R
EFERENCES
[1] I. Stoica, et al., “Chord: A scalable peer-to-peer lookup
protocol for internet applications,” IEEE/ACM Trans.
Networking, vol. 11, no. 1, 2003, pp. 17-32.
[2] H. Shen, Z. Li, and J. Li, “A DHT-aided chunk-driven overlay
for scalable and efficient peer-to-peer live streaming,
IEEE
Trans. on Parallel and Distributed Systems, vol. 24, no.11,
2013, pp. 2125-2137.
[3]
A.I.T., Rowstron, P. Druschel, “Pastry: Scalable,
decentralized object location and routing for large-scale
peer-to-peer systems,” Proc. of the 18th IFIP/ACM
International Conference on Distributed Systems
Platforms, 2001, pp. 329-350.
[4] P. Maymounkov, D. Mazieres, “Kademlia: a peer-to-
peer information system based on the XOR metric,” 1st
International Workshop on Peer-to-peer Systems, 2002.
[5] F. Picconi and L. Massoulie, “Is there a future for mesh-based
live video dtreaming?” Proc. 8th Int’l Conf. Peer-to-Peer
Computing (P2P), 2008, pp. 289-298.
[6] J. Venkataraman and P. Francis, “Chunky spread: Multi-tree
unstructured peer-to-peer multicast,” Proc. 5th Int’l Workshop
Peer-to-Peer Systems (IPTPS), 2006, pp. 1-10.
[7] F. Wang, Y. Xiong, and J. Liu, “mTreebone: A hybrid
tree/mesh overlay for application-layer live video multicast,”
Proc. 27th Int’l Conf. Distributed Computing Systems
(ICDCS), 2007.
[8] R. Sweha, V. Ishakian, and A. Bestavros, “AngelCast: Cloud-
based peer-assisted live streaming using optimized multi-tree
construction, Proc. 3rd Multimedia Systems Conference
(MMSYS), 2012, pp. 191-202.
[9] A. H. Payberah, H. Kavalionak, V. Kumaresan, A. Montresor,
and S. Haridi, “CLive: Cloud-assisted P2P live streaming,”
Proc. 12th IEEE Int’l Conf. Peer-to-Peer Computing (P2P),
2012, pp. 79-90.
[10] F. Wang, J. Liu, and M. Chen, “CALMS: Cloud-assisted live
media streaming for globalized demands with time/region
diversities,” IEEE Infocom, 2012, pp. 199-207.
[11] Y. Zhou, D. M. Chiu, and J. C. Lui, “A simple model for
analyzing P2P streaming protocols,” IEEE Int’l Conf.
Network Protocols (ICNP), 2007, pp. 226-235.
[12] S. Lan, S. Zhang, X. Zhang, and Z. Guo, “Dynamic
asynchronous buffer management to improve data continuity
in P2P live streaming, IEEE 3rd Int’l Conf. Computer
Research and Development (ICCRD), 2011, pp. 65-69.
41Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
49 / 79
An Intelligence System Based on Social Web Mining and Its Application in Health
Care in Hong Kong
Kin Keung Lai
1,2
1
Management Sciences
College of Business, City University of
Hong Kong, Hong Kong.
2
International Business School
Shaanxi Normal University
Xian, China
mskklai@cityu.edu.hk
Juan Shi
1,2
1
Management Sciences
College of Business, City University of
Hong Kong, Hong Kong.
2
Management Science and Engineering
Xi’an Jiaotong University
Xian, China
juanshi4-c@my.cityu.edu.hks
Gang Chen
The 705th Research Institute
China Shipbuilding Industry Corporation
Xi’an, China
james.gang.chen@gmail.com
Abstract—In China, two systems—Chinese Information system
for disease control and prevention (CDC) and the National
Adverse Drug Reaction (ADR) Monitoring System--have been
established for monitoring infectious diseases and for reporting
ADR events. However, surveys have shown that nontrivial
problems have affected the performance of these information
systems adversely. Specifically, the absence of reasonable work
and supervision systems, namely the systemic factors constitute
the majority of the reasons for underperformance of these
information systems. Also, not just some leaders but also quite
a few doctors do not give sufficient importance to reports of
infectious cases or ADR events, which is thought to be a waste
of manpower and financial resources. Nowadays social media
websites have attracted wide interest from both the academia
and the industry, because of its very large user base, updating
in real time, abundant incoming data and the fact that it offers
prolific information about consumers’ behavioral
characteristics. In this proposal, we plan to address the
aforementioned issues by implementing an Intelligence System
based on Social Web Mining (ISSWM). ISSWM can help
detect ADR events, assess the trends of infectious diseases,
probe people’s opinions towards some medical institutions or
on healthcare related issues in real-time, at a cost that is
extremely low compared with the existing systems.
Keywords-big data; Web 2.0; social web mining; opinion
mining; sentiment analysis; epidemics surveillance.
I. INTRODUCTION
The Chinese Information system for disease control and
prevention (CDC) was established in China for control and
prevention of infectious diseases in 2005. Medical and health
institutions at the grass roots level are required to report
cases of infectious diseases to this system, which helps the
concerned department(s) to keep abreast of the situation and
make timely decisions. Besides that, the National Adverse
Drug Reaction (ADR) Monitoring System has also been
launched to improve surveillance of ADR events. However,
there are some nontrivial problems in these information
reporting systems. For example, a survey, which investigates
the reports of infectious cases via CDC in 17 medical
institutions, has found some salient reasons that seem to be
responsible for underperformance of CDC. First, reports
using CDC have not been given sufficient importance by
staff of these medical institutions. Second, there is a lack of
effective work and monitoring mechanisms. Third, medical
staff involved in reporting via CDC change frequently and
lack adequate knowledge about infectious diseases [1]. As a
result, missing reports rate is estimated at 11.76% and the
rate of consistency between infectious diseases report cards
and information input into the system is only 78.25%.
Another survey about ADR has reached similar conclusions:
both information reliability and timeliness leave much to be
improved. From the above, it can be concluded that
management factor and human factor are two main reasons
of unsatisfactory performance of these information reporting
systems. There is a long way to go and a lot of efforts are
needed to address the problems of these systems.
Fortunately, we have entered the era of Web 2.0, which is
characterized by high degree of public engagement and
initiative. In China, the number of registered users of Sina
Weibo, the biggest social platform, has reached 600 million
and there are as many as 61.4 million daily active users [2].
It is worth noting that medical and health related data,
generated on social media, constitute an important part of
Big Data relevant to the medical area. Generated by users
directly, they are constantly updated and reflect every aspect
of people’s lives [3]-[4]. More importantly, data on the social
media are much easier to access via API compared with
databases owned by the government. The rapid growth of
online social networking sites and the public availability of
data have made analysis of social media data easy and
convenient [5]-[6].
This proposal envisages implementation of an
intelligence system based on social web mining (ISSWM),
which extracts insightful information from the unstructured
or semi-structured data available on social media to answer
questions related to people’s main concerns in healthcare
area at a certain moment. How is an epidemic spreading in
42Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
50 / 79
the country or within a certain area? What are people’s
specific expectations from some particular medical
institutions? However, it needs to be pointed out that the
purpose of ISSWM is not to substitute any existing
authoritative information-gathering websites, such as CDC
or the ADR monitoring system. Instead, ISSWM functions
as an intelligence system complementary to the existing
information systems. For example, it could serve as a plug-in
for the existing systems. The advantages of ISSWM over
existing information reporting systems, like CDC and ADR
monitoring system, are as follows:
1. Timeliness and low cost. ISSWM can automatically
collect, process and analyze data from social media websites
in real time, using computer algorithms. These features of
the proposed ISSWM have several implications. First,
manual effort is greatly reduced and the process virtually
costs nothing. Second, real-time data collection and analysis
helps decision-making department(s) stay up to date with the
latest situation. Third, ISSWM can effectively mitigate
problems resulting from lack of reasonable regulation in
CDC or the national ADR monitoring system and problems
which arise from medical staff’s lack of awareness regarding
the importance of reporting infectious cases.
2. Tremendous user base generated by integrating users
from multiple social networking websites. Since data can be
collected from numerous users of multiple social media
websites, ISSWM can efficiently extract information related
to infectious diseases, ADR events and other related issues in
healthcare sector. The very large number of users provides
the much needed credibility to results of ISSWM.
3. Use of big data. During events of public interest, such
as presidential debates, there are hundreds of thousands of
tweets per minute [7]. Interactions on websites, such as
Twitter and Sino Weibo, contain a huge amount of
meaningful information and this presents some challenge to
ISSWM, which is taken into account in design.
The rest of this paper is organized as follows. In Section
2, we discuss prior work and their limitations on exploiting
social media data in healthcare area. Section 3 presents the
framework of ISSWM. Section 4 concludes the paper.
II. LITERATURE REVIEW
Recently, a considerable amount of work has been
dedicated to exploiting data on Twitter, most of which has
focused on forecasting epidemics based on interactions on
Twitter [8]-[12]. For example, the public sentiment with
respect to H1N1 was tracked and actual disease activity was
measured on the basis of information embedded in the
Twitter stream [9]. Furthermore, Twitter data, such as
friendship relationships [3], the proximity of users’
geographical locations and users’ mobility, are used to
estimate the likelihood of whether a specific person is likely
to get infected [10]-[11]. Influenza-related messages are
identified by leveraging a document classifier and these
messages achieve a correlation of 0.78 with the Centers for
Disease Control and Prevention (CDC) statistics [12].
A recently emerging research area is to explore online
reviews to understand patients’ attitudes towards healthcare
services. For example, it has been revealed that there is
more than 80% agreement between patients’ own
quantitative ratings of care and those derived using sentiment
analysis, from online free-text comments on different aspects
of healthcare [13]. Reference [14] analyzed a corpus
consisting of nearly 60,000 reviews with a probabilistic
model of text. The output of the model was found to
significantly correlate with state-level measures of healthcare.
After reviewing these literature, we find out that there are
some limitations in the existing research. First, they focus on
data collected from only one social media website, Twitter,
as reported in [8]-[12], English National Health Service
website in [13], and RageMDs in [14]. Due to a single source
of data, it is doubtful whether the findings still hold in the
context of the whole society; Twitter users are not
necessarily representative of the wider population [15].
Second, existing research concentrates on solving one or two
specific problems, such as detecting flu trend or tracking
people’s sentiment regarding H1N1. An integrated system is
required to provide comprehensive knowledge about the
current status and sentiment related to healthcare issues in
the society. Third, most of prior studies have adopted a
methodology called post-collection analysis [8, 10-12]. That
is to say, the analysis depends on pre-retrieved tweets or a
well-prepared corpus [13-14], and there is a time delay that
cannot be negligible in their analysis results. Thus a real-time
system is in required to ensure online analytical processing
in a streaming fashion. In this way, the breakout of
emergency events can be detected almost instaneously and
the latest situations are reported in real time, given people’s
high engagement in social networking websites. Last but not
the least, China has a much larger population compared with
any other country, producing an unprecedented amount of
social media data. However, healthcare related data on
Chinese social networking websites are virtually untapped.
To bridge the abovementioned gaps, we propose an
intelligence system based on social web mining (ISSWM). It
collects data from multiple social networking websites,
operates incessantly (24 hours*7 days) and delivers valuable
results in real time. The structure of ISSWM is illustrated in
Figure 1. Extensive explanation about ISSWM is presented
in Section 3.
III. INTRODUCTION ABOUT ISSWM
The proposed ISSWM shall employ crawling techniques
and a variety of machine learning algorithms, such as
clustering, feature selection, k-nearest neighbors, regression
and so on, to obtain insightful knowledge from social media
websites. ISSWM mainly comprises four modules shown in
Figure 1.
1. Data collecting and preprocessing module
As mentioned above, ISSWM incessantly collects data
from several Chinese social networking websites, such as
Sina Weibo [16], Tencent Weibo [17], Renren Net [18] and
so on. Crawling technologies are also used as a
complementary method to retrieve data, as there are a few
limitations on obtaining tweets through API for some social
networking websites [19]. Besides that, API access is not
available for non-social networking websites [20]-[25]. It can
43Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
51 / 79
be expected that the raw tweets contain a lot of noisy data
such as wrong spellings, punctuation, images, or tweets that
have nothing to do with healthcare sector.
It is proposed to preprocess the raw data in the following
way. First, regular expressions are used to delete numbers,
punctuations, and non-Chinese characters. Second,
stopwords, which are meaningless words, are filtered out.
Third, we crawl healthcare websites, such as 120ask.com
[20], mingpaohealth.com [22], etc., in order to get a corpus
of healthcare related words. Inversed Term Frequency (IDF)
[26] is used to evaluate how important a given word is.
Words with higher IDF value are thought to be more related
with healthcare sector. To further improve the accuracy of
this corpus, a priori knowledge is used to filter out words
which get higher IDF values but seldom appear in healthcare
sector. After the healthcare corpus is established, raw tweets
obtained from social networking websites are examined
using this corpus. A tweet that does not share any word with
this corpus is treated as non-relevant and is screened out.
Clean tweets are integrated, sorted and converted into a
unified format, as illustrated in Table 1 and Table 2.
Consistency needs to be checked as the tweets come from
different data sources. Table 1 illustrates profile information
about each user and Table 2 shows the format of information
about each tweet. Relationship data such as each user’s
followers and friends are also kept in this step.
TABLE I. INFORMATION ABOUT USERS
Field Description
id ID of the user for whom to return results for table
screen_name
The screen name of the user for whom to return
results for
statuses_count The number of tweets issued by the user
description Description about the user himself/herself
friends_count The number of users this account is following
followers_coun
t
The number of followers this account currently has
location The user-defined location for this account’s profile
time_zone The time zone this user declares themselves within
… …
TABLE II. INFORMATION ABOUT TWEET
Field Description
id
The integer representation of the unique identifier for
this Tweet
text
The actual text of the status update
created_at Time when this Tweet was created
favorite_count
Indicates approximately how many times this Tweet
has been “favorited” by Twitter users
retweet_count Number of times this Tweet has been retweeted
place Indicates that the tweet is associated a place
user The user who posted this Tweet
2. Topic detection and classification module
We know that when the topic changes, keywords change
correspondingly, since frequently used terms differ greatly
across different areas. Latent dirichlet allocation (LDA)
algorithm, which can represent each tweet as a random
mixture of latent topics [27], is employed in this module to
detect hot topics in social networking websites.
Suppose we would like to know what are the Top N
topics attracting people’s attention at a given point of time.
First, we will retrieve tweets in the past month. After the
preprocessing procedure (Module 1), we obtain clean tweets
that focus on issues related to healthcare area, all of which
are used to develop a LDA model, with the parameter
“number of topics” set as N. Different weights are assigned
on different topics that a given tweet can possibly be related
to. The topic category on which a given tweet gets the
highest weight is used to classify this tweet. In this way, all
clean tweets are divided into different topic categories. Next,
the N topics are sorted in terms of the number of tweets
which belong to a particular topic. Finally, we get the Top N
hot topics in healthcare area. The topic with a higher rank
attracts more attention on social networking websites.
When a new unseen tweet arrives, it is transformed by
using the established LDA model and is assigned to the topic
category on which this tweet scores the highest. In this way,
we identify which topic is acquiring more and more attention
and, consequently, the trend of each of the Top N topics.
Combing users’ profile information and auxiliary
information of a tweet, we can gain an even deeper
understanding about the hot topics. For example, by
exploiting the geographic information, we get to know the
particular location where a specific topic induces a strong
response among the citizens. With users’ profile information,
we may understand which group of people are involved in a
particular topic. From time to time, we can check the content
of the top N topics and get to know the change in the hot
topics.
Like other hot topics, infectious diseases have their own
key words, which can be obtained by a priori knowledge or
learned by tweets on the internet. With these keywords,
relevant tweets are extracted from multiple social networking
websites and used to estimate the trend of some infectious
diseases. In order to detect ADR events, a specialized corpus
of medical-related words needs to be developed at the very
beginning and any tweet which contains a relevant medical
word or the name of a medicine in this corpus should be
given special attention.
3. Module of opinion mining and sentiment analysis
After preprocessing (Module 1) and topic detection
(Module 2), tweets are assigned to different topic classes. In
order to grasp people's opinions or sentiments on some given
topic, we use features like opinion words or phrases,
negations, emoticons, syntactic dependency and so on to
judge the sentiment underlying a tweet. Tweets which do not
have these features are treated as neutral tweets and are
discarded. Support vector machine (SVM) algorithm is
employed to classify opinions into negative or positive and
regression is used to identify the intensities of sentiments. As
44Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
52 / 79
both are supervised algorithm, labelled data are required to
develop the model. We produce labelled data by labelling the
polarities and intensities of the training tweets manually.
4. Opinion predicting module
Apart from classification of sentiments into positive or
negative, we are interested in predicting people’s opinions on
new events. This problem has two dimensions. First, we
want to forecast people’s opinions on an upcoming event.
Second, an event has occurred, but not everyone has publicly
expressed his or her opinion toward this event. Therefore we
would like to estimate the opinions of this silent group. We
claim that people with similar characteristics tend to hold
similar opinions toward the same event. Besides, people tend
to hold similar opinions toward events of the same nature.
The first hypothesis is validated in the following way.
First, we cluster people by characteristics, based on
their profile information using clustering algorithms. Profile
information generally includes age, occupation, gender and
location and so on, as illustrated in Table 1, which can be
extracted from tweets and people’s own descriptions about
themselves [28]. And then, we compare the similarity of
people’s attitudes towards a specified event, who are located
in the same cluster with similarity of people’s attitudes
towards the same event, who are randomly chosen from a
larger population. If the former similarity is significantly
larger than the latter similarity in a statistical sense, the first
hypothesis is validated. Following the same line, the second
hypothesis could also be tested.
Based on the two validated hypotheses, we can use
collaborative filtering which include memory-based
collaborative filtering or model-based collaborative filtering
to estimate people’s opinions on new events.
IV. CONCLUSION
This paper envisages a new application of social web
mining in medical and healthcare area. A system called
ISSWM is proposed to exploit and analyze raw tweets from
multiple social media websites. By ISSWM, we can predict
the trends of infectious diseases, understand and estimate
people’s attitudes towards healthcare related issues, detect
ADR events and stay informed about other healthcare related
hot topics. ISSWM makes the best of social media websites
and can deliver analysis results very efficiently at an
extremely low cost.
ACKNOWLEDGMENT
The work described in the paper is financially supported
by the Theme Based Research Grant, RGC of Hong Kong
(TRS Project 8770001).
REFERENCES
[1] Y. He, Y. d. Liu, F. l. Ma, F. k. Xing, “Sample analysis in 17
medical institutions on the reports of infectious cases using
Chinese Information system for disease control and
prevention ” Journal of Preventive Medicine of Chinese
People's Liberation Army, no. 02, 2014, pp. 125-126.
[2] Sina (2014) Sina Weibo is the best crowd funding platform,
take Dark Horse Games as an example.
http://tech.sina.com.cn/i/2014-02-26/11479193540.shtml.
Accessed 10 June 2015
[3] F. Griffiths, et,al, “Social networks–The future for health care
delivery,” Social science & medicine, vol. 75, no. 12, pp.
2233-2241, 2012.
[4] S. Brennan, A. Sadilek, H. Kautz, "Towards understanding
global spread of disease from everyday interpersonal
interactions," in Proceedings of the Twenty-Third
international joint conference on Artificial Intelligence, 2013,
pp. 2783-2789..
[5] C. McNab, “What social media offers to health professionals
and citizens,” Bulletin of the World Health Organization, vol.
87, no. 8, 2009, pp. 566-566.
[6] R. N. Rimal, and M. K. Lapinski, “Why health
communication is important in public health,” Bulletin of the
World Health Organization, vol. 87, no. 4, 2009, pp. 247-247a.
[7] M. A. Russell, Mining the Social Web: Data Mining
Facebook, Twitter, LinkedIn, Google+, GitHub, and More: "
O'Reilly Media, Inc.", 2013.
[8] R. Chunara, J. R. Andrews, J. S. Brownstein, “Social and
news media enable estimation of epidemiological patterns
early in the 2010 Haitian cholera outbreak,” The American
Journal of Tropical Medicine and Hygiene, vol. 86, no. 1,
2012, pp. 39-45.
[9] A. Signorini, A. M. Segre, P. M. Polgreen, “The use of
Twitter to track levels of disease activity and public concern
in the US during the influenza A H1N1 pandemic,” PloS one,
vol. 6, no. 5, 2011, pp. e19467.
[10] Sadilek A, Kautz HA, Silenzio V, Modeling Spread of
Disease from Social Interactions. In: ICWSM, 2012.
[11] Sadilek A, Kautz HA, Silenzio V, Predicting Disease
Transmission from Geo-Tagged Micro-Blog Data. In: AAAI,
2012.
[12] A. Culotta, "Towards detecting influenza epidemics by
analyzing Twitter messages," in Proceedings of the first
workshop on social media analytics, 2010, pp. 115-122..
[13] F. Greaves, D. Ramirez-Cano, C. Millett, A. Darzi, L.
Donaldson, “Use of sentiment analysis for capturing patient
experience from free-text comments posted online,” Journal
of medical Internet research, vol. 15, no. 11, 2013.
[14] B. C. Wallace, M. J. Paul, U. Sarkar, T. A. Trikalinos, M.
Dredze, “A large-scale quantitative analysis of latent factors
and sentiment in online doctor reviews,” Journal of the
American Medical Informatics Association, vol. 21, no. 6,
2014, pp. 1098-1103.
[15] D. King, D. Ramirez-Cano, F. Greaves, I. Vlaev, S. Beales, A.
Darzi, “Twitter and the health reforms in the English National
Health Service,” Health policy, vol. 110, no. 2, 2013, pp. 291-
297.
[16] Sina Weibo. "Sina Weibo API," 6th, July, 2015;
http://open.weibo.com/wiki/%E9%A6%96%E9%A1%B5.
[17] Tencent. "Tencent Weibo API," 6th, July, 2015;
http://dev.t.qq.com/.
[18] Renren. "Renren API," 6th, July, 2015;
http://wiki.dev.renren.com/wiki/API.
[19] Sina Weibo. "Rate limiting on Sina Weibo API," 6th, July,
2015; http://open.weibo.com/wiki/Rate-limiting.
[20] 120ask. "120ask," 6th, July, 2015; http://www.120ask.com/.
45Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
53 / 79
[21] haodf. "haodf," 6th, July, 2015; http://www.haodf.com/.
[22] mingpaohealth. "mingpaohealth," 6th, July, 2015;
http://www.mingpaohealth.com/cfm/Main.cfm.
[23] xywy. "xywy," 6th, July, 2015; http://www.xywy.com/.
[24] ewsos. "ewsos," 6th, July, 2015; http://www.ewsos.com/.
[25] 39net. "39net," 6th, July, 2015; http://www.39.net/.
[26] Wiki. "TF-IDF," 6th, July, 2015;
https://en.wikipedia.org/wiki/Tf%E2%80%93idf.
[27] D. M. Blei, A. Y. Ng, M. I. Jordan, “Latent dirichlet
allocation,” the Journal of machine Learning research, vol. 3,
2003, pp. 993-1022.
[28] J. Li, A. Ritter, E. Hovy, “Weakly supervised user profile
extraction from twitter,” ACL, Baltimore, 2014.
Raw data
Topic Detection
Data Collecting
Opinion analysis
Opinion analysis
Opinion analysis
Opinion analysis
……
Data set
about topic 1
Data set
about topic 2
Data set
about topic 3
Data set
about topic n
……
Predit public opinion
User profile
Predicting
result
Public opinion
about topic 1
Public opinion
about topic 2
Public opinion
about topic 3
Public opinion
about topic n
Module 1 Module 2 Module3 Module 4
Figure 1. ISSWM for health care.
46Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
54 / 79
Energy- and Priority-Aware Traffic Engineering for Content-Centric Networking
Ling Xu, Tomohiko Yagyu
Cloud System Research Lab
NEC Corporation, Japan
Email: lingxu@gisp.nec.co.jp, yagyu@cp.jp.nec.com
Abstract—Content-Centric Networking (CCN) is a new network
architecture aiming to solve many fundamental problems of
IP networks. We are interested in the application of CCN
carrier networks during natural disasters. During these times,
carrier networks need to save energy until reinforcement arrives.
Traffic delivery is prioritized, where high-priority traffic such
as disaster news receives superior transmission quality. Multiple
Tree-based Traffic Engineering (MTTE), a recently proposed
traffic engineering scheme, realized the goal of energy-saving by
forwarding traffic on minimal spanning trees. In addition, several
priority-aware transmission schemes were proposed recently.
However, no CCN communication scheme that realizes both
energy-saving and prioritized routing has ever been considered.
In this paper, we propose Priority-aware MTTE (PMTTE) to
fill this vacuum. Given a certain energy consumption limitation,
PMTTE optimizes the transmission quality of the high-priority
traffic first. Then, it uses the remaining energy consumption quota
to improve the low-priority traffic’s quality. We compared the
performance of PMTTE against priority queue-enabled MTTE
(as a naive energy- and priority-aware CCN traffic engineering
scheme). Simulation shows that compared with MTTE, PMTTE
can improve the performance of high-priority traffic by up to
80%, and of low-priority traffic by more than 30%.
Keywords–content-centric networking; priority-aware; energy
efficiency; multiple trees
I. INTRODUCTION
A. Content-Centric Networking
Content-Centric Networking (CCN) is a network architec-
ture that is proposed recently for solving IP architecture’s
problems [1]. Conventional IP networks focus on retrieving
needed contents from certain hosts. Nowadays, however, users
are only concerned about the contents, not from where to
obtain them. In CCN networks, each piece of content is given a
unique name. Users tell the networks the names of the needed
contents. The networks independently identify hosts providing
the contents and then fetch the contents to the users.
One of CCN’s main features is that routers carry caches
to reduce redundant traffic. Each host in the network may
provide some contents and this host is called the producer
of its contents. The producer splits each of its contents into
a set of chunks. The producer also registers the name of
each of its contents on each router - a process called content
publishing. Hosts that request the content is called the content’s
consumers. When a consumer needs a piece content, it sends
one request (called an Interest) for each chunk of the content.
The Interest is then forwarded to the producer according to
the information the producer registered on routers. When the
producer receives an Interest, it sends the requested chunk back
to the consumer. Each router along the chunk’s forwarding
route tries to store the chunk in its cache. The next time the
router receives an Interest for the same content, if the chunk
is still in its cache, the router replies directly with the chunk
to the consumer.
We are interested in CCN’s application in carrier networks
during natural disasters and hope to make the CCN network
both energy-efficient and priority-aware.
On one hand, networks consume a huge amount of energy,
and improving network energy efficiency has been a popular
research topic in recent years. In CCN networks, in-router
caches cost extra energy. Hence, the networks generally con-
sume more energy than conventional IP networks.
Xu et al. proposed Multiple Tree-based Traffic Engineering
(MTTE) to reduce CCN’s energy consumption [2]. Research
has found that network interfaces (shortened as faces hence-
forth) of routers in modern networks are generally underuti-
lized [3]. The idea of Xu et al. is to shut down as many faces of
routers as possible. MTTE splits traffic on multiple tree-like
networks that are generated based on the physical network.
The trees are generated in such a way that the number of
faces included in the trees is minimized. Since trees generally
contain fewer faces than the original physical network, energy
can be conserved.
On the other hand, real-world carrier networks deliver
traffic of different priorities. During natural disasters, disaster
alarms and news broadcasting are more important than en-
tertainment TV programs. When disasters occur, the network
traffic usually surges suddenly. MTTE is priority-agnostic and
cannot ensure the performance of the high-priority traffic.
Several priority-aware transmission schemes have been
proposed for CCN networks recently. However, to the best
of our knowledge, none of the existing proposals realizes both
energy- and priority-aware transmission. Through this paper,
we hope to fill this vacuum.
B. Research Goal
In this paper, we assume that the network delivers two
kinds of traffic: high-priority and low-priority. Examples of
the high-priority traffic include live streaming such as disaster
news. Such traffic is mission-critical and has to be delivered
to users stably before certain deadlines. For examples, during
the Great East Japan Earthquake [4], tsunami hit the seashore
several minutes after the earthquake. If tsunami news can arrive
at people in time, thousands of lives can be saved. The low-
priority traffic can be other types of communication like binary
file transmission. Compared with the high-priority traffic, low-
priority traffic is less urgent and can be delayed or dropped.
The simplest idea to implement priority-aware transmission
is to always deliver the high-priority traffic first using priority
47Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
55 / 79
queues. By “using priority queues,” we mean that each face
contains one high-priority queue and one low-priority queue.
High-priority (low-priority) packets are piled on the high-
priority (low-priority) queue in a first-in and first-out manner.
The face always sends packets in the high-priority queue before
forwarding any packets in the low-priority queue. However,
this approach would result in serious quality degradation of
the low-priority traffic.
Through this research, we aim to create a traffic engi-
neering scheme that provides these properties: (1) it reduces
energy consumption of the whole network system; (2) given a
upper bound for the system energy consumption and a lower
bound for the transmission quality of the high-priority traffic,
it guarantees both the bounds without greately sacrificing the
quality of the low-priority traffic.
C. Proposal Summary
When a consumer requests a piece of content, we use
session to denote the process in which the consumer receives
all chunks. Each Interest belongs to one certain session. We
call sessions transmitting high- (low-) priority contents the
high- (low-) priority sessions.
Sessions that have at least one chunk not returned before
a certain deadline φ
dead
are regarded as failed sessions. We
define Session Failure Rate (SFR) as the ratio of failed sessions
among all sessions. Session failure rate is used as the main
transmission quality metric in this paper.
We propose Priority-aware MTTE (PMTTE), and com-
pare the performance of PMTTE with priority queue-enabled
MTTE (as a naive priority and energy-aware CCN transmis-
sion scheme). Simulation shows that, compared with MTTE,
PMTTE reduces the session failure rate of high-priority traffic
by 80% and of low-traffic by 30%.
The rest of this article is organized as follows: Section
II highlights relevant, prior literature in this field; Section
III briefly introduces MTTE since MTTE is the base of
PMTTE; Section IV introduces PMTTE’s design and Section
V evaluates PMTTE’s performance; and Section VI is the
conclusion.
II. RELATED WORK
Several priority-aware transmission schemes have been pro-
posed for CCN networks recently. However, to the best of our
knowledge, no scheme that balances the trade-off between the
network energy consumption and the transmission quality of
prioritized traffic has been considered yet. Kim et al. proposed
reserving bandwidth for the high-priority traffic [5] where
their communication scheme is similar to the IntServ band-
width reservation scheme for IP networks [6]. During natural
disasters, important messages need to be delivered to more
people. Routers have limited capacity and cannot forward all
packets. Psaras et al. proposed that packets should be given
names that specify their priorities [7]. Routers independently
decide whether to forward a packet according to the routers’
remaining capacities and the packet’s priority. The objective
of this research is orthogonal from ours. Immediately after
a natural disaster, communication traffic may dramatically
increase and overload the network. Psaras et al. proposed pri-
oritizing contents according to the importance of the contents
to the contents’ consumers [8]. Somaya et al. used prioritized
queues for delivering packets with different deadlines [9].
Traffic is classified into multiple groups. Routers maintain
queues for accommodating different groups and packets with
near deadlines are delivered first. Tsilopoulos et al. realized
that for streaming contents in Information-Centric Networking
(synonym of CCN), it is inefficient for consumers to request
each chunk. They suggested that routers work as proxies that
request content for consumers. [10]
Research effort has been made in assessing CCN’s energy
efficiency using simulation [11][12]. Some research compared
the energy efficiency of CCN with existing IP-based content
delivery techniques such as content delivery networking and
peer-to-peer networking. Song et al. [13] noticed that in
modern carrier networks, a great amount of traffic is generated
from the edge. [13] uses GreenTE - an existing energy-aware
TE mechanism designed for IP networks [3] - for reducing
energy consumption of the core network, and uses CCN for
eliminating redundant traffic generated by the edge network.
However, Song et al.’s approach does not reduce the energy
consumption of CCN itself.
III.BACKGROUND KNOWLEDGE OF MTTE
In CCN, each router contains a Forwarding Information Base
(FIB). An FIB contains a set of entries and each entry is a
mapping from one prefix to one or multiple router face(s).
Suppose that the FIB contains two entries fe0=‘‘/Asia/’’:{
face2} and fe2=‘‘/Asia/Tokyo/’’:{face2,face5} and that an
incoming packet has a name “/Asia/Tokyo/music.mp3.” The
router finds the face whose FIB entry’s prefix matches the
packet’s name and forwards the packets via this face.
A. Tree-Based Congestion Control
In MTTE, the network contains a central server called the
controller. The controller maintains a database named the tree
set. Initially, the controller creates one spanning tree based on
the physical network topology and adds the tree into the tree
set. The controller sends the tree set to routers. Routers update
their respective FIBs so that they can forward the packets along
the trees.
When congestion is about to occur, the controller creates
more trees to spread the traffic and mitigate congestion. Specif-
ically, routers periodically report the utilization of their faces
to the controller. The controller calculates the Congestion Rate
(CR) as the maximal value of the received utilization. When
CR > φ
tcc
, where φ
tcc
is a system parameter, the controller
creates another tree and adds it to its tree set. This mechanism
is termed as Tree-Based Congestion Control (TCC).
B. Face Weight Calculation
We use ST s to denote the tree set in MTTE. Faces included
in the tree set are called live faces, denoted by E(ST s). Faces
not included the tree set are called free faces.
When the controller creates a new tree it calculates the
weight of each face. Based on the weights, the controller runs
Kruskal’s minimal spanning tree algorithm. The controller then
asks routers to evenly deliver traffic on trees in the new tree
set.
To reduce energy consumption, the new tree should not
dramatically increase the number of live faces. To this end,
we call the live faces with utilization higher (lower) than φ
ce
the congested faces (uncongested faces). The controller assigns
weights to faces so that the uncongested faces are chosen first,
free faces are chosen later, and congested faces are chosen last.
48Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
56 / 79
Namely, underutilized live faces are more likely to be used to
create the new tree, which suppresses energy consumption.
A critical design in MTTE is that the diameters of the
routing trees must be suppressed. Otherwise, when the diame-
ters are large, not only is the transmission delay increased but
also the cache utilization efficiency is decreased. Congestion is
more likely to occur and the transmission quality deteriorates
quickly.
MTTE uses a Betweenness-based heuristic algorithm to
suppress the diameters of the routing trees. For each face e, the
controller computes its “face betweenness” (e.be) - a widely
used metric in graph theory [14]. Informally, e.be represents
the number of the shortest paths in the entire network that
traverse e. Imagine that routers c and p are a pair of consumer
and producer, and sp is the shortest path between c and
p on G. Intuitively, if more faces with high betweennesses
are added to st, the probability that the data transmitted on
st between c and p are delivered along the shortest path is
higher. Accordingly, the diameter of st will be small. Based
on this observation, in MTTE, the controller selects faces
with higher betweennesses first. When the controller creates
the initial tree, for each face e, it calculates e.eb, and sets
e.weight = 1/e.eb. When subsequent trees are created, the
controller makes the weights of uncongested faces directly
proportional to the faces’ utilization and makes the weights of
free faces inversely proportional to the faces’ betweennesses.
The complete tree creation algorithm is shown in Algorithm
1.
C. Hash-Based Traffic Splitting
Each time the controller changes the tree set, it asks routers
to update their FIBs. First, the controller sends the tree set to
all routers. Suppose that H is a collision-proof hash function
preloaded on each router, and N (d) is the name of content
d. Routers update their FIBs so that d is forwarded on the
H(N (d))%K-th tree, where K is the size of the new tree set.
After the update, packets are generally evenly delivered on
each tree and the congestion can be mitigated.
D. Tree Removal
When CR < φ
uu
, where φ
uu
is a preloaded system
parameter, the controller removes the last tree in the tree set.
The controller sends the new tree set to all routers, and asks
routers to update their FIBs. Routers shut down their adjacent
faces that are not included in the new tree set. Accordingly,
energy consumption can be reduced.
IV. PMTTE: THE DESIGN
A. Problem Analysis
PMTTE uses priority queues-enabled MTTE as the base
architecture. This architecture has two problems:
First, priority queues alone are not sufficient to ensure
the transmission quality of the high-priority traffic. When
the network has bandwidth bottlenecks, high-priority packets
will drop anyway. MTTE works well in resolving bandwidth
bottlenecks by creating more trees before congestion really
occurs. Hence, in MTTE, the main reason that routers drop
packets can be that the traffic changes so fast that no sufficient
trees are created. Certainly we can accelerate tree creation by
using smaller φ
tcc
. However, creating more trees consumes
more energy.
1: E
UE
= Faces in E(ST s) with utilization <= φ
ce
.
2: E
CE
= Faces in E(ST s) with utilization > φ
ce
.
3: E
free
: Faces not in E(ST s)
4: U
max
= maximum of face utilization of E
UE
5:
6: The controller calculates the betweenness e.eb of each face
e.
7:
8: if |ST s| = 0 then
9: // The controller is creating the initial tree
10: for all e in the network do
11: e.weight = 1/e.eb
12: end for
13: else
14: // The controller creates a new tree for mitigating
congestion
15: for all e ∈ E
UE
do
16: e.weight = e.utilization
17: end for
18: for all e ∈ E
free
do
19: e.weight = U
max
+ 1/e.eb
20: end for
21: for all e ∈ E
CE
do
22: e.weight = U
max
+ 2
23: end for
24: end if
25:
26: The controller generates a minimal cost spanning tree st
using Kruskal’s algorithm on the whole network.
27: if st is different from all the existing trees then
28: return st
29: else
30: return FAILED
31: end if
Figure 1: Based on current face utilization, the controller
generates a new spanning tree (Algorithm 1 from [2]).
Second, in MTTE, routing trees are created in such a
way that they heavily overlap each other for reducing energy
consumption. Since priority queues aggressively drop low-
priority packets, the transmission quality of the low-priority
traffic will be heavily impacted when congestion occurs.
We propose Priority-Dependent Routing (PDR) and Face
Separation (FS) to address these problems. Algorithm 2 lists
the PMTTE algorithm’s work flow and functionalities.
B. Priority-Independent Routing
We assume that each packet of the high- (low-) priority
traffic contains an attribute tag that specifies content’ priorities,
as done in [7]. Routers can know the priorities of incoming
packets by checking priority tags.
To improve the transmission quality of the high-priority
traffic, we propose PDR. PDR forwards the high- and low-
priority traffic on independent trees. We use whole traffic to
denote the total of the high- and low-priority traffic. We call
the trees used for forwarding the high- (low-) priority traffic
the H (L) trees. Analogously, MTTE only has one tree set (for
the whole traffic) and we call it the W trees. We call the three
parameters required by MTTE, φ
tcc
, φ
ce
and φ
uu
, a threshold
set. The controller holds independent threshold sets for the
high- and low-priority traffic, respectively. The φ
tcc
of the H
trees is generally set lower than φ
tcc
of the L trees. Hence, the
49Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
57 / 79
high-priority traffic will have more sufficient routes and is less
likely to get congested. Meanwhile, since we only accelerate
the H trees’ creation, energy consumption can be suppressed.
PDR improves the quality of the low-priority traffic as well.
As traffic is separately forwarded, the congestion among traffic
of different priorities in routers’ forwarding queues can be
mitigated. As a result, transmission quality of both the high-
and low-priority traffic can be improved.
1) Implementation
To implement PDR in PMTTE, each router holds two
FIBs - one for the high-priority traffic and the other for the
low-priority traffic. When the system launches, the controller
creates two spanning trees and adds them into the H trees and
the L trees, respectively. These two trees are created using the
original MTTE tree-creating algorithm. Each time the H trees
(L trees) change, the controller sends the changed tree set to
routers. When a router receives the H (L) trees, it updates
its high- (low-) priority FIB using the similar traffic splitting
algorithm of MTTE. The difference with MTTE is that, in
MTTE, all prefixes (both high- and low-priority) are evenly
split on the W trees. In PMTTE, the high- (low-) priority
prefixes are evenly split over the H (or L) trees.
For each face e, we use U
e
(H), U
e
(L) and U
e
(W ) to
denote the utilization caused by the high-priority traffic, the
low-priority traffic and the whole traffic, respectively. Similar
to MTTE, routers in PMTTE periodically send the utilization
of their faces to the controller. The difference between MTTE
and PMTTE is that in MTTE, routers only send the U
e
(W ).
In contrast, in PMTTE, routers send both U
e
(H) and U
e
(W ).
Based on the received utilization, the controller adds trees
when congestion is about to occur (Algorithm 2). We use the
congestion rate of the high- (low-) priority traffic, denoted as
θ(H) and θ(L), as the maximum of the received U
e
(H) and
U
e
(W ). When θ(H) (θ(L)) exceeds φ
tcc
(H) (φ
tcc
(L)), the
controller creates a tree for the H trees (L trees). Namely,
the low-priority traffic flows in PMTTE in the same way the
whole traffic flows in MTTE. Note that θ(L) is the maximum
of U
e
(W ) instead of U
e
(L). The latter means that when the
controller creates a new tree for the L trees, the controller is
more likely to pick faces used by the H trees. Later, when
these faces get congested, low traffic’s quality will seriously
suffer.
2) Judgment of Live Faces
In the original tree creation algorithm, the controller needs
to judge whether a face is live (Line 18, Algorithm 1). In
MTTE, all faces used in the tree set are regarded as live faces.
In contrast, in PMTTE, faces used in the H trees (rather than
used in both the H trees and the L trees) are regarded as
live faces when the controller creates a new tree for the high-
priority tree set (Algorithm 6). The reason is as follows.
Recall that the transmission quality is closely affected
by the diameters of the routing trees (Section III-B). Like
in MTTE, when the controller in PMTTE creates a tree
and needs to choose a free face, it chooses faces of higher
betweennesses first in order to reduce the tree’s diameter (Line
12, Algorithm 5). Accordingly, trees created later generally
have larger diameters than trees created earlier.
The L trees are created based on the maximum of the whole
traffic. Naturally, the number of trees in L tree set is likely to
be larger than the number of trees in the H tree set. On average,
trees in the L tree set are likely to have larger diameters than
trees in the H tree set. By only using faces in the H tree set
as live faces, PMTTE can suppress the diameter of the new
high-priority tree.
3) Removing Underutilized Trees
If θ(H) (or θ(L)) is lower than φ
uu
(H) (or φ
uu
(L)), the
controller removes the last tree from the H (or L) tree set.
C. Face Separation
Due to the overlapping nature of the tree creation algorithm
of MTTE, the H trees and L trees would still heavily cover
each other, which threatens the low-priority traffic’s quality. To
address this problem, when the controller creates a new low-
priority tree, it increases the weights of faces that are used in
the H trees by a factor of φ
F S
(Algorithm 4). The controller
can explicitly specify the extent that the L trees should disjoint
from the H trees. This technique is called FS, and φ
F S
is
named the Face Separation Rate.
1: Creates the initial H tree;
2: Creates the initial L tree;
3: while 1 do
4: if θ(H) > φ
tcc
(H) then
5: CreateTree(H)
6: end if
7: if θ(L) > φ
tcc
(L) then
8: CreateTree(L)
9: end if
10: Sleeps for a certain period T
update
;
11: end while
Figure 2: PMTTE work flow
1: weights = {};
2: for all e in the network do
3: weights[e] = GetF aceW eight(e, newT reeT ype);
4: end for
5: Calculates a Kruskal minimal spanning tree using
weights;
6: Adds the new tree to the newT reeT ype trees;
7: Sends the newT reeT ype trees to routers for updating the
FIBs;
Figure 3: The CreateTree(newT reeT ype) method.
newT reeT ype is ‘H’ or ‘L’ when the new tree to create is
for the H (or L) trees.
1: weight = GetDefaultWeight(e, newT reeT ype)
2: if newT reeT ype is H then
3: return weight
4: else
5: return weight · φ
F S
6: end if
Figure 4: The GetFaceWeight(e, newT reeT ype) method for
FS. The controller calculates the weight of a face e when
creates a new tree.
D. Energy Consumption and Transmission Quality Trade-off
We use session failure rate (defined in Section I-C) as the
metric for transmission quality, and use Live Face Rate (LER)
as the metric for energy consumption. We use SFR(H) and
SFR(L) to denote the SFR of the high- and low-priority traffic,
respectively. LER is defined as γ/|E|, where γ is the number
50Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
58 / 79
1: if IsLiveFace(e, newT reeT ype) then
2: u = GetUtilization(e, newT reeT ype);
3: // e is an uncongested face
4: if u < φ
ce
(newT reeT ype) then
5: weight = u;
6: else
7: // e is a congested face
8: weight = u + 2;
9: end if
10: else
11: // e is a free face
12: weight = φ
ce
(newT reeT ype) + 1/e.betweenness;
13: end if
14: return weight;
Figure 5: The GetDefaultWeight(e, newT reeT ype) method.
1: if newTreeType is H then
2: return e is used in H trees;
3: else
4: return e is used in H trees or L trees;
5: end if
Figure 6: The IsLiveFace(e, newT reeT ype) method
of faces used in the L and H trees, and |E| is the total number
of faces in the physical network.
During disasters, network administrators determine a upper
bound, φ
energy
, for the system energy consumption and a
lower bound, φ
sfr
, for the transmission quality of the high-
priority traffic. The administrators dynamically adjust φ
tcc
(H)
and φ
tcc
(L) in a heuristic manner (Algorithm 8). When
real system energy consumption exceeds φ
energy
, φ
tcc
(L) or
φ
tcc
(H) is increased to reduce the system energy consumption.
When LER <= φ
energy
and SF R(H) > φ
sfr
, φ
tcc
(H) is
reduced to improve the high-priority traffic’s quality. When
LER <= φ
energy
and SF R(H) < φ
sfr
, φ
tcc
(L) is reduced
to improve the low-priority traffic’s performance. In this way,
administrators can guarantee the requestments for energy con-
sumption and the high-priority traffic, and minimize the low-
priority traffic’s quality degradation.
V. EVALUATION
This section evaluates PMTTE’s performance by comparing
PMTTE with priority queue-enabled MTTE (MTTE).
A. Topology
Our simulation is performed on ndnSIM – a simulation
platform developed by UCLA for CCN-related research [15].
The simulation runs on the network topology of au-
tonomous system 3257 (AS3257). This topology is provided in
Rocketfuel network dataset [16], a dataset that has been used in
network research [17][18]. Each node in AS3257 represents a
router. We extract the largest connected component of AS3257
and use all the remaining nodes for creating trees. AS3257
contains three types of routers: cores, gateways and leaves.
1: if newT reeT ype is H then
2: return U
e
(H);
3: end if
4: return U
e
(W );
Figure 7: The GetUtilization(e, newT reeT ype) method.
1: while 1 do
2: if LER > φ
energy
then
3: if φ
tcc
(L) < 1.0 then
4: Increases φ
tcc
(L);
5: else
6: if φ
tcc
(H) < 1.0 then
7: Increases φ
tcc
(H);
8: end if
9: end if
10: else
11: if SF R(H) > φ
sfr
then
12: if φ
tcc
(H) > 0 then
13: Reduces φ
tcc
(H);
14: end if
15: else
16: if φ
tcc
(L) > φ
uu
then
17: Reduces φ
tcc
(L);
18: end if
19: end if
20: end if
21: Sleeps for T
update
;
22: end while
Figure 8: Energy consumption and transmission quality
trade-off
According to the definition of Rocketfuel datasets, leaves are
the routers with degrees equal to or less than two, gateways are
the routers directly connected to the leaves, and the remaining
routers are cores.
We assume that in real-world CCN networks, consumers
are adjacent to leaves; low-priority producers, such as servers
of entertainment applications, are adjacent to gateways and
leaves; high-priority producers, such as servers of national TV
stations, connect to cores. We assign one low-priority producer
to each leaf and gateway, and assign one high-priority producer
to each core. In total, 132 low-priority producers and 108
high-priority producers are located. Each producer provides
20 different pieces of content.
B. Simulation Parameters
We assign ConsumersP erN ode high- and low-priority
consumers to each leaf. Each high- (low-) priority consumer
randomly chooses one high- (low-) priority content from the
whole high- (low-) priority prefix pool. End consumers start
requesting selected content after certain delays so that network
traffic volume follows a sine wave shape (for simulating the
traffic spike during the disaster).
We change ConsumersP erNode to control the
maximal network traffic volume. Two volumes, with
ConsumersP erN ode being 60 and 80, are evaluated. The
number of high-priority consumers is 4,800 (60 x 80) and
6,400 (80 x 80), respectively. Similarly, the number of
low-priority consumers is 4,800 and 6,400 as well. Each
consumer issues 10 Interests per second (10 ips), meaning
that maximally 10 chunks should be returned by certain
producers per second. The size of each chunk is 1024 bytes.
According to the simulation, the average number of hops
between each pair of consumer and producer is approximately
10. Hence, maximally 983 MB (4,800 x 2 x 10 ips x 10 hops
x 1024 bytes) traffic can flow within the network per second
when ConsumersP erNode = 60. Meanwhile, each face
has a capacity of 5 Mbps. The network contains 420 faces.
51Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
59 / 79
The theoretical capacity of the whole network is 260 MBps
(5 Mbps x 420 / 8). Therefore, we believe that the network
is sufficiently congested to evaluate the effectiveness of the
proposed algorithm when ConsumersP erNode = 60 and
80.
The threshold set of the L trees in PMTTE is equal to
the threshold set of the W trees in MTTE. Both sets have
φ
tcc
= 0.8, φ
ce
= 0.6 and φ
uu
= 0.5.
The number of high- and low-priority consumers is equal.
Intuitively, the high- and low-priority traffic have equal vol-
umes. As φ
tcc
of MTTEs’ W trees is 0.8, φ
tcc
of PMTTE’s
H trees should be equal or less than half of it, i.e., 0.4. For
PMTTE’s H trees, we set the default values of φ
tcc
, φ
ce
and
φ
uu
to 0.4, 0.3 and 0.15, respectively.
C. Performance Metrics
We evaluate the performance using SFR and LER. Session
failure rate is directly decided by the session failure deadline
φ
dead
. φ
dead
should be set according to the mean round-trip
time (RTT) of the system. The RTT in different communication
systems may vary greatly. In [19], the authors evaluated the
Quality of Experience (QoE) of a streaming system and they
assumed a RTT of 150ms. Streaming applications generally
request low RTTs. In this paper, we are considering general
carrier networks where both streaming and non-streaming
traffic exist. Hence, the RTT in our system should be larger
than the 150ms. In our simulation, we set φ
dead
to 200ms.
Recall that LER is defined as γ/|E|. For MTTE, γ is the
average number of faces used in the whole tree set during the
simulation, For PMTTE, γ is the average number of faces used
in the L and H trees during the simulation.
We expect to see that, compared with MTTE, PMTTE
further reduces SFR(H) and SFR(L) at the price of a mod-
erate increase in LER. The effectiveness of PDR and FS are
evaluated, respectively.
We repeat each evaluation scenario ten times and display
the mean values of each performance metrics.
Parameters used in the simulation are summarized in Table
I.
TABLE I: Simulation parameters
Parameter Value
Total number of faces 420
Total number of routers 240
Number of gateway routers 52
Number of leave routers 80
Number of core routers 108
Prefixes per producer 20
Chunk size 1024 bytes
Cache size 2000 chunks
(For PMTTE) Default H φ
tcc
, φ
ce
, φ
uu
0.4, 0,3, 0.15
(For PMTTE) Default L φ
tcc
, φ
ce
, φ
uu
0.8, 0.6, 0.5
(For MTTE) Default Whole φ
tcc
, φ
ce
,
φ
uu
0.8, 0.6, 0.5
Duration of one simulation run 50 seconds
Update interval T
update
2 seconds
High-priority producer location cores
Low-priority producer location gateways,
leaves
Session failure deadline φ
dead
0.2 seconds
D. Evaluating Priority-Independent Routing
Figure 9: Session failure rate for the low priority traffic.
Legends indicate the algorithm and traffic volume used for
evaluation. For example, “pmtte, 60” indicates that PMTTE
is evaluated when ConsumersP erNode = 60.
Figure 10: Session failure rate for the high priority traffic.
Figures 9 and 10 compare the session failure rates of the
low-priority traffic and high-priority traffic between PMTTE
and MTTE. On the Figures, φ
tcc
(H) increases from 0.15 to
0.4 on the x-axis. For evaluation displayed in this section, φ
F S
is set to 1.0 (i.e., FS is disabled).
[19] shows that in video streaming systems, the users’ QoE
degrades rapidly when the packet drop rate exceeds certain
thresholds. Since we consider prioritized transmission, instead
of limiting the absolute value of the session failure rate, it
is more important to focus on the performance superiority of
the high-priority traffic over the low-priority traffic. In this
paper, we assume that SFR(H) should be lower than 2% to
guarantee decent QoE. Figure 10 shows that when φ
tcc
(H) is
0.15, SFR(H) can be reduced to lower than 2%.
As φ
tcc
(H) decreases, more H trees are created and
SFR(H) of PMTTE is suppressed. In Figure 10, it is no-
table that PDR effectively reduces SFR(H) when φ
tcc
(H) is
reduced to 0.15. SFR(H) reduces from 7.2% to 1% when
52Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
60 / 79
Figure 11: Live face rate.
ConsumersP erN ode = 60, and reduces from 8.3% to 1.5%
when ConsumersP erNode = 80. Meanwhile, when SFR(H)
is less than 2%, SFR(H) can be than more than 10 times lower
than SFR(L) (Figure 9 and Figure 10).
Meanwhile, as expected in Section IV-B, PDR remarkably
reduces SFR(L) as well (Figure 9), as compared with MTTE.
When ConsumersP erNode = 60, SFR(L) decreases from
51% to 33%. When ConsumersP erNode = 80, SFR(L)
decreases from 56% to 34%.
Figure 11 plots the influence of PDR on energy consump-
tion. Clearly, the improvement in transmission quality is at
the price of higher energy consumption. Even though PMTTE
still suppresses LER to be lower than 40%. In real-world
network systems, the network operator can tune φ
tcc
(H) to
make the trade-off between transmission quality and energy
consumption.
E. Evaluating Face Separation
Figure 12: Session failure rate for the low priority traffic.
Legends indicate the algorithm and traffic volume used for
evaluation. For example, “pmtte, 60” indicates that PMTTE
is evaluated when ConsumersP erNode = 60. (FS enabled)
Figures 12 and 13 compare the session failure rates be-
tween PMTTE and MTTE when FS is enabled. The results are
Figure 13: Session failure rate for the high priority traffic.
(FS enabled)
Figure 14: Live face rate. (FS enabled)
obtained with φ
tcc
(H) = 0.15 (the value obtained in Section
V-D where SFR(H) of PMTTE falls below 2%). The face
separating rate φ
F S
increases from 1.0 to 5.0 on the x-axis.
When φ
F S
takes 1.0, the results are equal to those obtained
when FS is disabled and PDR only is enabled.
Figures 12 and 13 clearly reveal that priority queues
deteriorate the low-priority traffic. Meanwhile, FS can ef-
fectively mitigate this deterioration. As the face sepa-
ration rate increases, SFR(L) decreases (33% to 27%
for ConsumersP erNode = 60, and 34% to 32% for
ConsumersP erN ode = 80). Moreover, FS does not affect
SFR(H) (Figure 13), meaning that network operators can
improve the performance of the low and high traffic using
PDR and FS independently.
Figure 14 shows that FS moderately increases energy con-
sumption. However, LER is still lower than 50% in PMTTE.
VI.CONCLUSION AND FUTURE WORK
Through this research, we aim to create a traffic engineering
scheme that is both energy- and priority-aware for CCN
networks. The challenge is to avoid heavily impacting the
transmission quality of the low-priority traffic. Our proposal,
53Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
61 / 79
PMTTE, splits the high- and low-priority traffic on separate
routing trees. This enables us to flexibly increase the tree
recreation frequency for the high-priority traffic to suppress
the high-priority traffic congestion. At the same time, this
change mitigates the collision between the high- and low-
priority traffic, which improves the quality of both. Last but
not least, by aggressively separating the routes of the high-
and low-priority traffic, we can actively improve the quality
of the low-priority traffic. Simulation using a real-world ISP
network topology shows that compared with naive priority
queue-enabled MTTE, PMTTE can boost the quality of both
the high- and low-priority traffic by up to 50%. As for future
work, we plan to improve the transmission quality of the
high-priority traffic by creating routing trees that minimize the
distances between the high-priority producers and consumers.
Meanwhile, MTTE is a centralized scheme and will fail when
the connection between routers and the control is cut. We plan
to make MTTE more fault-tolerant, especially during disasters.
ACKNOWLEDGMENT
The work described in this paper was, in part, performed in
the context of the FP7/NICT EU-JAPAN GreenICN project.
REFERENCES
[1] L. Zhang et al., “Named data networking (NDN) project,” Relat
´
orio
T
´
ecnico NDN-0001, Xerox Palo Alto Research Center-PARC, 2010.
[2] L. Xu and T. Yagyu, “Multiple-tree based online traffic engineering
for energy efficient content centric networking,” in Proc. The Sixth
International Conference on Advances in Future Internet (AFIN2014).
IARIA, 2014, pp. 121–127.
[3] M. Zhang, C. Yi, B. Liu, and B. Zhang, “GreenTE: Power-aware
traffic engineering,” in Network Protocols (ICNP), 2010 18th IEEE
International Conference on. IEEE, 2010, pp. 21–30.
[4] N. Mimura, K. Yasuhara, S. Kawagoe, H. Yokoki, and S. Kazama,
“Damage from the great east japan earthquake and tsunami-a quick re-
port,” Mitigation and Adaptation Strategies for Global Change, vol. 16,
no. 7, 2011, pp. 803–818.
[5] Y. Kim, Y. Kim, and I. Yeom, “Differentiated services in named-data
networking,” in Computer Communications Workshops (INFOCOM
WKSHPS), 2014 IEEE Conference on. IEEE, 2014, pp. 452–457.
[6] B. Braden, D. Clark, and S. Shenker, “Integrated services in the internet
architecture: an overview,” Internet Requests for Comments, RFC
Editor, RFC 1633, June 1994, http://www.rfc-editor.org/rfc/rfc1633.txt.
[Online]. Available: http://www.rfc-editor.org/rfc/rfc1633.txt
[7] I. Psaras, L. Saino, M. Arumaithurai, K. Ramakrishnan, and G. Pavlou,
“Name-based replication priorities in disaster cases,” in Computer
Communications Workshops (INFOCOM WKSHPS), 2014 IEEE Con-
ference on. IEEE, 2014, pp. 434–439.
[8] G. Tyson, E. Bodanese, J. Bigham, and A. Mauthe, “Beyond content
delivery: Can icns help emergency scenarios?” Network, IEEE, vol. 28,
no. 3, 2014, pp. 44–49.
[9] S. Arianfar, P. Sarolahti, and J. Ott, “Deadline-based resource manage-
ment for information-centric networks,” in Proceedings of the 3rd ACM
SIGCOMM workshop on Information-centric networking. ACM, 2013,
pp. 49–54.
[10] C. Tsilopoulos and G. Xylomenos, “Supporting diverse traffic types in
information centric networks,” in Proceedings of the ACM SIGCOMM
workshop on Information-centric networking. ACM, 2011, pp. 13–18.
[11] N. Choi, K. Guan, D. C. Kilper, and G. Atkinson, “In-network caching
effect on optimal energy consumption in content-centric networking,” in
2012 IEEE International Conference on Communications (ICC). IEEE,
2012, pp. 2889–2894.
[12] U. Lee, I. Rimac, D. Kilper, and V. Hilt, “Toward energy-efficient
content dissemination,” Network, IEEE, vol. 25, no. 2, 2011, pp. 14–19.
[13] Y. Song, M. Liu, and Y. Wang, “Power-aware traffic engineering with
named data networking,” in Seventh International Conference on Mobile
Ad-hoc and Sensor Networks (MSN), 2011. IEEE, 2011, pp. 289–296.
[14] M. Girvan and M. E. Newman, “Community structure in social and
biological networks,” Proceedings of the National Academy of Sciences,
vol. 99, no. 12, 2002, pp. 7821–7826.
[15] A. Afanasyev, I. Moiseenko, and L. Zhang, “ndnsim: Ndn simulator for
ns-3,” Named Data Networking (NDN) Project, Tech. Rep. NDN-0005,
Rev, vol. 2, 2012.
[16] “Rocketfuel dataset,” https://github.com/cawka/
ndnSIM-ddos-interest-flooding/tree/master/topologies/rocketfuel
maps cch, (retrieved: July, 2015).
[17] N. Spring, R. Mahajan, and D. Wetherall, “Measuring ISP topologies
with rocketfuel,” ACM SIGCOMM Computer Communication Review,
vol. 32, no. 4, 2002, pp. 133–145.
[18] C. Yi et al., “A case for stateful forwarding plane,” Computer Commu-
nications, 2013, pp. 779–791.
[19] T. N. Minhas, O. Gonzalez Lagunas, P. Arlos, and M. Fiedler, “Mobile
video sensitivity to packet loss and packet delay variation in terms of
qoe,” in Packet Video Workshop (PV), 2012 19th International. IEEE,
2012, pp. 83–88.
54Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
62 / 79
On Security-Effective Mobility-QoS Management Scheme in Heterogeneous Mobile
Networks
Hyeungwoo Lee
Network Development Group
SAMSUNG SDS Co. Ltd.
Suwon, Gyeonggi-do, 443-822, Republic of Korea
e-mail: hw77.lee@samsung.com (zen016@naver.com)
Jae-Young Choi and Jongpil Jeong
College of Information & Communication Engineering
Sungkyunkwan University
Suwon, Gyeonggi-do, 440-746, Republic of Korea
e-mail: {jaeychoi, jpjeong}@skku.edu
Abstract—To support efficient mobility, host-based mobility
management protocols have been developed. The
Authentication, Authorization, Accounting, and Charging
(AAAC) system is used in this paper to analyze the
effectiveness of the existing Proxy Mobile IPv6 (PMIPv6) and
Fast Handover for PMIPv6 (FPMIPv6) network security.
Furthermore, the IPv6 Mobility Management Protocol (MMP)
features, performance, and seamless transfer performance in
terms of packet loss probability are also analyzed. Our scheme
can be efficiently used to integrate Quality of Service (QoS)
and mobility to manage and control resources using a QoS
Broker (QoSB). The evaluation results show a better overall
performance for the fast handover structure of mobility
management techniques. PMIPv6 and FPMIPv6 are, in many
respects, the most efficient structures possible. Specifically, the
fast handover structure of the network-based mobility
management schemes shows the best results.
Keywords—Mobility-QoS; Security-Effective; Mobility
Management Protocol; PMIPv6.
I. INTRODUCTION
The wireless mobile environment is rapidly growing in
the digital environment that leads by human hands. The
Mobility Management Protocol (MMP) is a core protocol of
the wireless mobile environment. Mobile social networking,
computing, shopping, and so on will be achieved using the
mobility operating system. Various MMPs have been
developed for various mobility services. Particularly at the
network layer, mobility support techniques have been
developed by the Internet Engineering Task Force (IETF).
The Mobile IPv6 (MIPv6) specification was proposed, then
the Fast Handovers for MIPv6 (FMIPv6) and Hierarchical
MIPv6 (HMIPv6) specifications were developed as
extensions. As MIPv6 was developed, analysis of IPv6 MMP
was used to improve performance [1].
When host-based MMPs operated within the wireless
mobile telecom infrastructure, the telecoms companies and
technical developers became aware that it was not a suitable
solution for mobile services, especially for service providers,
as it was necessary to equip a Mobile Node (MN) with
mobility support inside the network protocol stack. Therefore,
MNs had to be upgraded or developed. This increased the
construction costs and complexity of the MN. Host-based
MMP has led to a lack of complex control operators. A new
approach to mobility services was required.
The extended protocol of Proxy MIPv6 (PMIPv6), Fast
Handover for PMIPv6 (FPMIPv6), has improved the
transmission rate by reducing transmission latency and
packet loss. In contrast to host-based MMP, network-based
MMPs (such as PMIPv6 or FPMIPv6) are in the early stages
of development. Improving the security of personal
authentication using the FPMIPv6 by applying the
Authentication, Authorization and Accounting (AAA)
mechanism has been studied. When moving between
domains in management, AAA techniques for authenticating
the MN are required [2]. The AAA scheme for the various
wired and wireless services performs authentication,
authorization, and billing. Today, many techniques in
conjunction with the AAA protocol are being investigated to
perform the functions of MN AAA, which is the
authentication process between the MNs. For instance, Zhou,
H et al. [3] proposed an FPMIPv6-based authentication
technique. When the MN enters a new network, this
technique protects the authenticating MN from security
threats, such as Replay Attack or Key Exposure.
In this paper, we propose a security-effective mobility
management scheme for IPv6-based networks using a
Quality of Service Broker (QoSB). This protocol can be
efficiently used to integrate Quality of Service (QoS) and
mobility to manage and control resources using a QoS
Broker (QoSB). The time latency is not significantly affected
because of the addition of the QoSB. In PMIPv6 and
FPMIPv6, the new proposal performs better with respect to
the handover latency, packet loss, and handover blocking
rates than the traditional MIPv6 scheme. These results are
shown for PMIPv6 and FPMIPv6 on a network security
system that uses Authentication, Authorization, Accounting,
and Charging (AAAC). Furthermore, in this paper, we
propose a unified criterion to analyze both host-based and
network-based MMPs.
This paper is organized as follows. Section 2 discusses
related work, and Section 3 describes the operating
procedures of the proposed scheme. In Section 4, a
performance evaluation of the proposed method is presented.
Finally, Section 5 presents some conclusions regarding these
results.
II. RELATED WORK
The PMIPv6 domain structure is composed of a Local
Mobility Anchor (LMA), Mobile Access Gateway (MAG),
55Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
63 / 79
and MN. An LMA is one kind of Home Agent (HA) that
serves as an MN in PMIPv6. In detail, a Home Network
Prefix (HNP) is allocated to the MN that maintains the
address and location information of all the MNs within the
domain and also ensures connection. The MAG is
responsible for network connectivity and routing functions
on behalf of the MN. It also performs MN mobility signaling
by tunneling through the LMA [4]. FPMIPv6 is a mobility
protocol that reduces the handover latency and packet loss
found in PMIPv6. It may reduce the loss of buffered packets
by creating a bi-directional tunnel between the previous
MAG (pMAG) and new MAG (nMAG) before making the
link-layer handover. And it consists of two modes: predictive
and reactive [5] [6] [7].
The QoS architecture is easily able to support end-to-end
QoS in terms of the operator. When the MN is moving, it is
guaranteed end-to-end connectivity and user maintenance.
The architecture is designed to control the scalable
deployment of resources in the access network. The core aim
of the architecture is the simultaneous support of mobility
and QoS. The QoS frameworks of various IETFs have
considered both purposes before the final design of the QoS
architecture. The advantages and disadvantages of
Integration Services (Intserv) [8] and Differentiated Services
(Diffserv) [9] have been discussed widely and are well
known. However, they do not specifically support mobility
and no hybrid solution integrates mobility and QoS.
Therefore, the aim of user mobility suggests an innovative
use of QoSB-related Fast HandOver (FHO) that includes a
Diffserv system to efficiently control and manage the
resources available [10]. This architecture is based on the
concept that the user is authorized by the service provider of
a contract. The QoSB, according to the user's agreement, is
responsible for resource allocation in an individual
subscription service. As proposed in [11], these services are
generally fixed transmission services (for example, a
“Guaranteed Rate of 64 KB/s,” or “Target Rate of 32 KB/s”)
but are equipped with a potential mechanism for flexible
service negotiation. The QoSB can manage the flow of
resources in the core network. To reduce the signal overhead,
the system is designed for a user/terminal so that it is not
necessary to explicitly reserve or release resources. The
services are requested by a simple Diffserv Code Point
(DSCP) marking to the outgoing packet. If the MAG
receives a packet from a particular user’s DSCP value, it
sends the required QoSB configuration. The QoSB
configures the MAG to fit the appropriate QoS policy based
on the information about the user. Services are implicitly
suspended by an inactivity timeout. This concept is
explained in more detail in [11].
III. PROPOSED SCHEME
The AAAC architecture is shown in Figure 1. It is based
on the AAA architecture, which is optimized for IPv6
enhanced auditing, metering, and charging. Considering that
AAAC is used for QoS in a PMIPv6 environment, the
architecture is designed to offer new functionality and
optimize the performance of the overall system.
Figure 1. Enhanced Generic AAA Architecture That Supports QoS-
enabled Mobility Management.
This architecture enables the subsequent auditing of
AAAC using the AAAC audit trail and other factors. Hence,
the policy repository is considered as a part of the policy-
based AAAC system. The AAAC system supports multiple
interfaces. AAAC performers can be treated with MN and
interfaces. Communication is performed by the User
Registration Protocol (URP). The Application Specific
Module (ASM) communicates with the QoSB. The
advantage of ASMs is additional flexibility, as various
service equipment can be easily processed using the same
method from the point of view of the AAAC system. ASM
uses the AAAC protocol to communicate with the AAAC
system and equipment-specified protocol to communicate
with service equipment. There are clear differences in the
services provided to the user in this architecture (e.g., QoS is
possible and the charging system follows the AAAC
requirements). For users previously provided and connected
via the ASM and extended AAAC protocol, on the other
hand, it is possible, if necessary, to communicate directly
with the AAAC system using dedicated communication.
AAAC system communication may be enhanced through an
appropriate expansion by the DIAMETER default protocol.
A key element of QoS service and charging is the means to
measure the service used. In an IP-based measurement
framework of the IETF Working Group, a variable was
defined for the IP flow that depends on the needs of the
network administrator. In the IPv6 network, the usage is
measured according to the type of service subscribed to by
the user and is sent to the QoS [2].
Figure 2. Registration Process.
56Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
64 / 79
For network operation and control, three steps may be
specifically identified. 1) Registration: in this architecture, an
MN/user can start using network resources after
authentication and authorization, just as in today's networks.
2) Acceptance: users should be allowed to use a specific
service prior to release by the network. 3) Handover: user
mobility should preserve existing resources when
transferring from one MAG to another.
The registration process to support end-to-end QoS is
shown in Figure 2. The registration process begins after the
CoA obtains the MN through automatic configuration with
the two-layer identifier. When a Duplicate Address
Detection (DAD) is performed, the uniqueness of the
registered address is checked. When obtaining a non-
authorized CoA, the user is authorized to consume only
enough resources for the registration message. However,
emergency calls can be made. However, as shown in Figure
2, the MN must start the authentication procedure by sending
user authentication information (message 1) to the MAG for
network connection. The request for this MAG is transmitted
to the controlled AAAC system (message 2). For more
complicated roaming, domain A (the AAAC domain) sends
a registration request to the home AAAC (message w) of the
MN. Domain A plays the role of a foreign domain that must
contact the home AAAC of the MN. The AAAC checks first
if there is a formal contractual relationship between the
management domain and its own management domain
(corresponding to the roaming agreement) as per the request.
If the result is affirmative, the home AAAC performs
authentication by verifying the provided credentials. The
home AAAC sends a request to the user’s LMA (message y).
Finally, the home AAAC responds to the AAAC of domain
A. A positive response consists of a user profile that contains
the information necessary to provide the requested service in
the foreign domain. The user profile contains the central
management profile, including all relevant user-specific
information related to the service provider. From a NVUP
(network view of the user's profile), a part of the profile is
sent from the AAAC server and it is necessary to provide the
requested service in the foreign domain. The user profile
contains the central management profile, including all
relevant user-specific information related to the service
provider. From a NVUP (network view of the user's profile),
a part of the profile is sent from the AAAC server to the
QoSB (message 3b) that also performs a DAD at this point.
The other set of profiles is sent to the AAA Attendant in the
MAG (message 3a). Along with the measurement and
security information that is delivered to the AAA attendant,
the NVUP includes all the required information related to
network services. Further, the AAAC informs the MN that
the registration is successful via the MAG (messages 3a and
4). After that, the MAG starts a task for the user and informs
the AAAC (message 4a). Accordingly, the authentication
phase is finished, and a user can access the network.
Figure 3 presents the process for granting authorization to
each network service (messages 5–11). The MN sends a
packet (message 5) with a DSCP code to request a specific
subscription service (e.g., 256 KB/s for priority network
access). One of the trailer packets, depending on the
configuration of the MN, may be a dummy packet or a
packet with real information. If the requested service does
not comply with the policy that has already been set in the
MAG, the MAG sends a request to QoSB via the QoS
manager. According to the analysis, the user’s NVUP, and
the availability of resources for the request, the QoSB
determines whether a message (message 7) is sent to the
MAG. The QoS manager of the MAG sets (message 7a) the
appropriate policy for the MAG, user and MN services, or
notifies the user of the service denial (message 7b). After
message 7a, the packet is sent to the MN that matches the
configured policy rules (message 8). Packets that have
different DSCP codes are subjected again to authentication.
When the packet reaches the final domain with other users, it
starts another QoS authorization process (message 8a). The
QoS manager of the MAG sends a policy question to the
QoSB (messages 9 and 10). If the QoSB has the resources,
the QoSB manager receives a positive response and is
configured for the MAG and its policies (message 10a). If it
does not have the resources, the MAG sends a reject message
regarding the service (message 10b). After message 10a, the
next packet to meet the policy is able to arrive at the other
terminal (message 11). In this way, two kinds of access
networks can provide the QoS level of an agreement. The
core network is monitored for performance (the end-to-end
QoS) as expected.
Figure 3. Authorization process.
One of the most difficult problems of IP mobility is
ensuring a constant level of QoS. As shown in Figure 4, for
user mobility in the network, the handover and network
messages are exchanged between QoSBs using the FHO
technique. When the MN begins to receive a weakened
signal from the current MAG (message 1), it sends and
receives the AS (Attendant Solicit), AA (Attendant
Advertisement), AReq (Authentication Request) messages,
and the handover procedure from the old MAG to a
neighboring “new MAG” is started. The MN builds its own
CoA and starts the handover process by sending an IP-
handover request for the new MAG through the old MAG
(message 4). The FHO module of the old MAG requests the
FHO Module of the new MAG and submits it to the QoS
manager module. The QoS manager immediately sends a
request to the existing QoSB (message 5). The previous
QoSB sends the handover request comprising the user’s
NVUP and a list of current user services to the new QoSB
(message 6). By default, this task is transferred to a new
57Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
65 / 79
QoSB in the context of the existing QoSB. The new QoSB
uses this information to check the availability of resources.
The MN sends the message to determine whether to perform
a handover to the QoS manager of the new MAG (message
7). This mechanism enables the QoSB to stop the handover
because of QoS constraints (e.g., the loss of bandwidth
resources). If the handover is possible, the QoS function
manager sends this information to the FHO module (message
7a) and configures the new MAG to accommodate the
moving MN. The new MAG starts the accounting process in
the user's AAAC system account (messages 8 and 11). To
complete the handover, the MN sends the LMA binding
update (messages 9 and 10) to the LMA. In addition, the
FHO module sends a handover response to the FHO module
of the old MAG (message 12). The new MAG begins bi-
casting. The MN handover to the new MAG is complete. If a
handover is completed within a QoSB domain, the QoSB for
controlling both MAGs is the same, and message 6 is not
sent. The remaining messages are the same. Information-
related security is exchanged between QoSBs in a similar
way.
Figure 4. Handover with QoS for End-to-End QoS Support.
IV. PERFORMANCE ANALYSIS
This section evaluates the performance of the proposed
method and the existing mobility management protocols.
A. Network Modeling
Figure 5 is shown a generic network topology model.
Figure 5. Network Model.
In Figure 5, the following hop count parameters are
defined for describing particular paths between
communication entities.
h
C−H
: It is the average number of hops between the
correspondent node (CN) and the HA.
h
C−G
: It is the average number of hops between the CN
and the gate.
h
H−G
: It is the average number of hops between the HA
and the gate.
h
G−A
: It is the average number of hops between the gate
and the MAG.
h
A−A
: It is the average number of hops between the
neighbor MAGs.
h
A−M
: It is the average number of hops between the
MAG and the MN.
The latency of registration lasts from when the user turns
on a device until it becomes available for use. There are two
types of latency. “Low-layer” latency refers to the delay
when connecting to a technology after the device is ready for
use. “High-layer” latency refers to the delay between sending
message 1 and the arrival of message 4 (Figure 2).
Re _gistration Delay LowLayerDelay HigherLayerDelay
First, this technique does not depend on a particular
network protocol or architecture. Second, it is dependent on
the link speed. If the user is roaming, it is the “electronic
distance” between the outside and home AAACs. The link
latency that occurs between the MAG and AAAC system is
small enough to be negligible, as it is normally the case that
the management infrastructure overprovisions the link to the
resource. The processing latency in the system occurs when
there is an overload in the number of requests or a database
is processing more requests that its capacity. However, in the
actual production of the network, this is sufficiently possible
to prevent using an appropriate computing or routing tool.
Thus, if a user is roaming, the limiting factor is the distance
between the external and home AAACs. In this case, the
latency is determined by the registration time.
Session setup latency is the time required for the user to
access the network. In Figure 3, the session setup latency is
58Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
66 / 79
the delay between messages 5 to 11. This latency is
composed of the processing time of the MAGs and QoSB,
indicating the distance between the MAGs, and the link
latency.
Handover latency can be an important parameter,
depending on a user's sensitivity. The handover method
should be quick as possible to provide seamless service to
users. The handover latency is composed of transmission
latency, computational latency, and two-layer handover
latency.
__
_
2_
Handover Delay Transmission Delays
Computation Delays
Layler Handover
The transmission latency is the sum of the latency that
occurs during MN-MAG, MAG-QoSB, QoSB-QoSB, and
MAG-MAG communication. The global handover latency is
the delay between messages 1 and 13, as shown in Figure 4.
However, the time during no connectivity or when user
terminal is not assigned resources is spent in the L2 handover.
The handover operation is simulated to determine mobility
seamlessness. After execution, measured values proving the
architectural concept is obtained that showed reduced global
handover latency, as well as low packet loss. The global
handover latency affects the relationship between cell
coverage of radio (range) and the speed at which the user can
move, therefore affecting the cellular network plan.
Figure 6. Timing Diagram for PMIPv6 Handover.
Figure 6 shows a timing diagram of a PMIPv6 handover.
Here,
( 6)PMIPv
HO
L
is defined to be the handover latency of
PMIPv6 and is expressed as follows.
( 6)
2
PMIPv
HO L LMA
L T T
Furthermore,
LMA
T
refers to the time spent exchanging the
PBU/PBAck messages between the MAG and LMA. During
the time required to send the RS Message, the LMA receives
the packet that is first sent.
( ) ( , )
LMA wl RS wd PBU G A lma packet
T d L d L h d
Variable
lma packet
d
denotes the time required for the first
data packet to be sent from the LMA to the MN. Because it
can be implemented by a static tunnel, a bi-directional tunnel
between the LMA and MAG is not necessary. When the
LMA receives a valid PBU message from the MAG, it sends
a data packet and PBAck message to the MN.
( ) ( , )
lma packet wl D wd D T G A
d d L d L L h
Latency
T
L
is contained in the
wd
d
account because the
data packet sent to the MN is tunneled between the LMA and
MAG. This account is included in the total. The data packet
is sent to the MN because of the tunnel between the LMA
and MAG. This is different than in HMIPv6. Even if
PMIPv6 and HMIPv6 send a similar message to the MN, the
PMIPv6 reduces the packet transmission overhead of the
wireless link. The FPMIPv6 has a concept similar to FMIPv6
and is composed of predictive and reactive modes.
Let
()
HO
L
be the handover latency in the mobility
management protocol that was developed in the previous
section. The () protocol is used as the indicator,
()
[]
HO
EL
is the
average value of
()
HO
L
, TR is the residence time on the
network, and its probability density function is denoted by
()
R
ft
. Here,
()
HO
L
is assumed to be exponentially distributed
by the accumulation function
()
()
T
Ft
. Hence,
()
HO
L
is the
element blocking the handover, and the handover block
potential
b
is expressed as follows.
()
()
()
()
0
Pr( )
[]
(1 ( )) ( )
1 [ ]
b HO R
c HO
T R u
c HO
LT
EL
F u f u d
EL
where
c
is the percentage of networks passing through
a boundary of the MN. If the MAG coverage area is circular,
c
is calculated as follows [12].
2
c
R
where
is the average speed of the MN and R is the
radius of the MAG coverage area.
B. Numerical Results
Performance analysis is used with the following system
parameter values:
CH
h
=4,
CH
h
=6,
HG
h
=4,
CA
h
=4,
AM
h
=1, E(S)=10,
=20ms, n=3,
f
L
=19bytes,
wl
D
=[10,40]ms,
wired
D
=0.5ms,
wired
BW
=100Mbps,
2L
T
=45.33ms,
DAD
T
=1000ms.
In this analysis,
f
ranges from 0 to 0.7 in increments of
0.05. Figures 7 shows the comparison of the handover
latency. The maximum value of
f
increases the probability
of an error in a wireless link when a packet is transmitted.
The number of retransmissions of the mobility signal
increases and results in an increase of the handover latency.
59Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
67 / 79
As shown in Figures 7, the handover latency of each
mobility management protocol is proportional to
f
.
Figure 7. Handover Latency (
f
,
wl
D
=10ms).
Figure 8 (left) shows the transmission failure probability
for
. As
increases, the MN must change rapidly. This
means that the MN is required to complete the transfer of a
high value in a shorter time. Therefore, as
increases, the
transfer failure rate of the mobility management protocol
also increases. In this analysis environment, if
is as high as
30 m, giving a handover probability of less than 0.05, only
two predictive high-speed transport protocols, FMIPv6 and
FPMIPv6, were able to function. Similar to the previous
results, MIPv6 handover probability block performance was
poor. This effect is notable as
increases. As seen in Figure
8 (right), most of the mobility management protocols are
influenced by R. However, the performance of predictive
FMIPv6 and FPMIPv6 is not affected. As with the results
shown in Figures 8, the handover latency for predictive
FMIPv6 and FPMIPv6 is short enough to avoid problems
caused by
or R.
Figure 8. Handover Blocking Probability versus
(left) and R (right).
V. CONCLUSION
In this paper, we integrate QoS and mobility to control
and manage the available resources effectively. This scheme
has the advantage that the time latency is very small because
of the added QoSB node. As shown in the results, the
existing mobility protocols and the proposed scheme are
analyzed with respect to handover latency, packet loss, and
handover blocking probability in networks based on PMIPv6
and FPMIPv6. The evaluation results show a better overall
performance for the FHO structure of mobility management
schemes that is equally applicable in a network-based
mobility management scheme. We can conclude that
PMIPv6 and FPMIPv6 are the most efficient structures in
many ways. In future, this approach will be applied to a
variety of service platforms, such as Internet of Things (IoT)
and verified in practical environments. We also plan to
continue expanding its research scope.
ACKNOWLEDGMENT
This research was supported by Next-Generation
Information Computing Development Program through the
NRF funded by the Ministry of Science, ICT & Future
Planning (NRF-2014M3C4A7030503) and Basic Science
Research Program through the NRF funded by the Ministry
of Education (NRF-2010-0024695).
REFERENCES
[1] J. H. Lee, J. M. Bonnin, and T. M. Chung, "Comparative Handover
Performance Analysis of IPv6 Mobility Management Protocols."
IEEE Tranaction on Industrical Electronics, Vol. 60, No. 3, 2013, pp.
1077-1088.
[2] V. Marques, X. P. Costa, and R. L. Aguiar, "Evaluation of a Mobile
IPv6-Based Architecture Supporting User Mobility QoS and AAAC
in Heterogeneous Networks," IEEE Journal on Selected Area in
Communication, Vol. 23, No. 11, 2005, pp. 2138-2151.
[3] Zhou, H., Zhang, H. and Qin, Y, "An authentication method for proxy
mobile IPv6 and performance analysis," Security Comm. Networks,
2009, pp. 445–454. doi: 10.1002/sec.83.
[4] H. W. Ko and J. P. Jeong, "dMMS: A Novel Distributed Dynamic
Mobility Management Scheme for Minimizing Signaling Costs in
Proxy Mobile IPv6 Networks," The Journal of The Institute of
Webcasting, Internet and Telecommunication, Vol. 12, No. 4, 2012,
pp. 65-80.
[5] S. H. Han and J. P. Jeong, "Intelligent Hierarchical Mobility Support
Scheme in F-PMIPv6 Networks," The Journal of the KICS, Vol. 38,
No. 4C, 2013, pp. 337-349.
[6] K. S. Go, U. S. Jung, and Y. S. Mun, "An Enhanced Fast Handover
for Proxy MIPv6 Scheme for Efficient Mobile Environment of The
Future Network," Journal of the institute of electronics engineerings
of Korea, Vol. 48, No, 1, 2011, pp. 84-91.
[7] H. Yokota, K. Chowdhury, and R. Koodli, "Fast Handovers for Proxy
Mobile IPv6," IETF RFC 5949, September 2010.
[8] R. Braden, D. Clark, and S. Shenker, "Integrated services in the
Internet architecture: An overview," IETF, RFC 1633, June 1994.
[9] D. Black, M. Carlson, E. Davies, Z. Wang, and W. Weiss, "An
architecture for differentiated services," IETF RFC 2475, December
1998
[10] C. Beaujean, N. Chaher, V. Marques, R. L. Aguiar, C. García, J. I.
Moreno, M. Wetterwald, and T. Ziegler, "Implementation and
evaluation of an end-to-end IP QoS architecture for networks beyond
3rd generation," in IST Mobile Summit, 2003, pp. 221–226.
[11] V. Marques, A. C. Casado, J. I. Moreno, and L. Rui, "A simple QoS
service provision framework for beyond 3rd generation scenarios," in
Proceeding of 10th International Conference on Telecommunications,
2003, pp. 1475–1481.
[12] R. Hsieh, Z.-G. Zhou, and A. Seneviratne, "S-MIP: A seamless
handoff architecture for mobile IP," in Proceeding of INFOCOM,
2003, pp. 1774-1784.
60Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
68 / 79
Delay Constrained ARQ Mechanism for MPEG Media Transport Protocol Based
Video Streaming over Internet
Hong-rae Lee, Tae-jun Jung, Kwang-deok Seo
Division of Computer and Telecommunications Engineering
Yonsei University
Wonju, South Korea
e-mail: kdseo@yonsei.ac.kr
Chang Ki Kim
Smart TV Media Development Team
Electronics and Telecommunications Research Institute
Daejeon, South Korea
e-mail: ckkim1@etri.re.kr
Abstract—MPEG Media Transport (MMT) is a new
international standard aiming at addressing the emerging
multimedia services over heterogeneous packet-switched
networks including Internet and broadcasting networks.
Due
to the heterogeneous characteristics of the broadcast and
broadband networks, MMT provides an efficient delivery
timing model to provide inter-network synchronization,
measure various kinds of transmission delays and jitter caused
by the transmission delay, and re-adjust timing relationship
between the MMT packets to assure synchronized playback.
Exploiting the delivery timing model, it is possible to
accurately estimate round-trip time (RTT) experienced during
MMT packet transmission. Based on the measured RTT, we
propose an efficient delay constrained ARQ (Automatic Repeat
reQuest) scheme which is applicable to MMT protocol based
real-time video communication over IP networks.
Keywords-video communication over Internet; ARQ scheme;
error control processing; MPEG Media Transport (MMT).
I. INTRODUCTION
In the recent years, digital broadcasting services and IP-
based multimedia services over the Internet including
mobile Internet have started being integrated and converged
[1]. With this trend, there have been so many changes in the
multimedia service environments such as media content
delivery networks, diverse video signals, 4K/8K video
transport systems, and various client terminals displaying
multi-format signals. It has become clear that the MPEG
standard has been facing several technical challenges due to
the emerging changes in those multimedia service
environments [2]. So, to address these technical challenges
to existing and emerging MPEG standards, ISO MPEG has
developed an MPEG-H standard suite (ISO/IEC 23008) for
the delivery of audio-visual information compressed with
high efficiency over heterogeneous environment. MPEG-H
suite consists of three functional areas: High Efficiency
Video Coding (HEVC) [3], 3D audio, and MPEG Media
Transport (MMT) [4].
In order to deploy efficient solutions for the transport of
HEVC video in an interoperable fashion, especially given
the recent increased demand of multimedia delivery in the
heterogeneous network environment, MPEG has launched a
new standardization work item, called MMT since the
middle of 2010. MMT has been working on addressing
technical challenges of existing standards due to recent
changes of multimedia delivery and consumption
environments and new requirements from emerging use
cases and application scenarios in the area of multimedia
services [5]. MMT aims to overcome current limitations of
available standards for media streaming by addressing
streaming format that is transport and file format friendly,
cross-layer optimized between video and transport layer,
error resilient for MPEG streams, convertible between
transport mechanisms and content adaptation to different
networks [6]. The challenge to error control schemes for the
real-time video communication is focused on how to
endeavor to recover the packet loss and then to reduce
impairment to the playback quality [7]. Many different error
control techniques have been proposed to solve these issues
and Automatic Repeat request (ARQ) has been known as
one of the promising solutions [8].
MMT provides delivery timing model as a means to
calculate jitter and the amount of delay introduced by the
underlying delivery network, so that constant delay for data
stream can be achieved [4]. Using the delivery timing model,
an MMT receiving entity provides information in the
feedback to allow the MMT sending entity to calculate the
RTT. To circumvent the packet loss problem, MMT employs
ARQ function as a basic error control technique in which the
MMT receiving entity asks for retransmission when packet
loss is detected at the receiver side [6][7]. When the RTT is
so high that the retransmitted packet will not arrive in time,
the retransmission will not improve the quality. The current
ARQ function of MMT, however, could result in many late
packets that arrive after play-out deadline because it does not
consider round-trip delay caused for the retransmission. To
circumvent this problem, we propose an efficient delay
constrained ARQ scheme for MMT packet based real-time
video communication.
The remainder of this paper is organized as follows. In
Section II, we overview MMT technology and summarize
the basic ARQ process supported in MMT. In Section III, we
describe the proposed delay-constrained ARQ scheme. The
experimental results and performance evaluation are
presented in Section IV. Finally, concluding remarks are
provided in Section V.
61Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
69 / 79
II. OVERVIEW OF MMT
In order to support efficient delivery and effective
consumption of coded media data for multimedia services
over packet-switched networks including IP networks and
digital broadcasting networks, MMT defines three
functional areas: encapsulation functional area, delivery
functional area, and signaling functional area as illustrated
in Figure 1 [4].
The encapsulation functional area defines the logical
structure of media content, the Package, and the format of
the data units to be processed by an MMT entity and their
instantiation with ISO base media file format (ISOBMFF)
as specified in ISO/IEC 14496-12. It produces Media
Processing Unit (MPU) as an output.
The delivery functional area defines an application layer
transport protocol including the payload format required for
transferring encapsulated media data from one network
entity to another. The payload format is defined to enable
the carriage of encoded media data which is agnostic to
media types and encoding methods.
The signaling functional area defines formats of signaling
messages to manage delivery and consumption of media
data.
Figure 1. MMT functional areas and interface
The MMT technology adopts ARQ technique as an error
control method for data transmission over error-prone
networks. Like the general behavior of the ARQ scheme
using a Negative Acknowledgment (NACK) [8], in case of
packet loss, a NACK is sent back from the receiving entity
to the sending entity and the sending entity retransmits the
lost packet [6].
Figure 2 illustrates the basic operation of the MMT ARQ
process [4]. The basic operation of ARQ process is as
follows. The first step is the generation of ARQ
configuration (AC) message by MMT sending entity, then it
is delivered to the receiving entity. MMT receiving entity
stores the ARQ configuration information. MMT receiving
entity continues with receiving MMT media packets and
checks whether there is a lost packet or not.
Figure 2. Operation of MMT ARQ process
Once MMT receiving entity determines that a packet has
been lost, an ARQ feedback (AF) message is generated
according to the configuration information defined in the AC
message. The AF message is sent to MMT sending entity
which will generate the retransmission packet to be sent to
MMT receiving entity. MMT receiving entity is able to
substitute the lost MMT packet with the retransmission
packet.
III. P
ROPOSED DELAY CONSTRAINED ARQ MECHANISM
FOR
MMT PROTOCOL
One of the well-known problems in ARQ happens if
networks experience severe congestion. In this situation, the
retransmission packet leads more congestion and causes the
further network degradation. Furthermore, in case of real-
time video communication, the ARQ will be successful only
if the retransmitted packet is received before its arrival
deadline. When the RTT is so high that the retransmitted
packet will not arrive in time, the retransmission will not
improve the quality [7].
The objective of the proposed delay constrained ARQ is
to suppress retransmitting requested lost packets that will
not arrive in time for playback. By using the delay
constrained retransmission, the MMT sending entity can
avoid unnecessary retransmission of the out-of-date packets
and therefore minimize the probability of wastefully
retransmitted packets. This results in a reduced amount of
data traffic wastefully injected into the network. The
proposed delay constrained ARQ uses arrival deadline to
62Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
70 / 79
decide whether MMT sending entity retransmits the
requested lost packet or not.
Figure 3 shows an exemplary timing diagram of the
proposed delay constrained ARQ. In Figure 3, arrival
deadline denotes the maximum tolerable latency for the
requested retransmission packet to arrive at the receiver. If
the retransmitted packet arrives later than arrival deadline,
even if the packet arrives intact, it is regarded as useless and
discarded so that decoding may not occur. Thus, the
recovery latency for the lost packet should be taken into
account for the real time video delivery wherein late repair
becomes useless because of the real time nature of the data.
RTT information which is crucial for the delay constrained
retransmission is already available to the sender by the
reception quality feedback (RQF) message supported in the
MMT standard. This could be achieved by using the
timestamp field existing in the MMT packet header. This
timestamp field specifies the time instance of MMT packet
delivery based on Coordinated Universal Time (UTC),
which corresponds to the sending time at the first byte of
MMT packet.
Figure 3. Timing diagram of the proposed delay-constrained ARQ
Based on the above analysis, when the MMT receiving
entity detects the loss of packet, the receiver sends AF
message to the MMT sending entity, and then the sender
takes the decision procedure for delay constrained
retransmission as shown in Figure 4 to decide whether to
transmit the requested lost packet or not. The estimated RTT
can be obtained a priori at the sending entity by RQF
message provided in MMT [6]. For more accurate
estimation of the RTT, up-to-date ARQ feedback delay (D
f
)
value can be informed to the sending entity by enclosing
ARQ feedback timestamp (T
ARQ
) into the AF message. ARQ
feedback timestamp corresponds to the time instant of
sending the AF message to the MMT sending entity. Using
the ARQ feedback timestamp, T
ARQ
, the sending entity can
obtain up-to-date ARQ feedback delay (D
f
). And this
updated D
f
value can be used to compute up-to-date RTT.
Finally, if the RTT is estimated to be less than the arrival
deadline, sender retransmits the requested lost packet to the
receiver. Otherwise, the sender decides not to transmit the
requested lost packet to the receiver. In order to implement
the proposed delay constrained ARQ scheme, the ARQ
feedback timstamp and the arrival deadline information
needs to be included in the AF message of MMT. The AF
signaling format supporting the proposed delay constrained
ARQ scheme could be found in [9] and [10], and has been
approved to be included in the standard document of MMT
AMD1 (Amendment 1) in the MPEG meeting.
Figure 4. Procedure to decide whether to retransmit the lost packet at the
MMT sending entity
The arrival deadline can be obtained at the receiver side
by considering the remaining amount of safely arrived
packets in the receiver buffer when the ARQ feedback
message is prepared for sending. Therefore, we exploit the
remaining amount of safely arrived packets in the receiver
buffer and average bit rate of the MMT packet stream when
the ARQ feedback message is prepared for sending.
63Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
71 / 79
IV. EXPERIMENTAL RESULTS
To evaluate the efficacy of the proposed delay
constrained ARQ scheme, we performed extensive
experiments over the MMT protocol-based HEVC video
streaming system. The video streaming system consisted of
an MMT sending entity and receiving entity. Before
initiating video streaming, the AC message which includes
the retransmission policy to be adopted by the MMT
sending and receiving entities in the event of packet loss is
sent to the receiving entity by the ARQ policy manager. The
MMT sending entity retains an MMT packet in the buffer
until the timeout, and it is thus available for retransmission.
If MMT sending entity receives an AF message requesting
retransmission of the lost packet, it checks whether the lost
packets is available in the retransmit buffer and it
retransmits the packet only if the arrival deadline value is
greater than RTT. To verify the performance of the
proposed ARQ scheme in an error-prone video transmission
environment, the NIST-Net [11] Linux-based network
emulation tool was used to emulate the packet-loss network
environment. We used the Stockholm video sequence with
HD 1080P resolution. The Stockholm video sequence was
coded by an HEVC encoder, which is being increasingly
used for broadcasting and mobile multimedia applications.
The run-time of the generated video streaming was 4 min 39
s, and the number of generated MMT packets to be
delivered was 193,042 in total. The size of each MMT
packet was 1500 bytes, and the receiving buffer was set to
store 300 MMT packets.
In the first experiment, the network delay was set to 50
ms while the packet loss rate has varied from 3% to 10% by
the NIST-Net network emulator. Figure 5 shows the ratio of
the number of renounced retransmission to the total number
of retransmission request for various packet loss rates. We
can observe that the ratio of the renounced retransmission
increases as the packet loss rate increases. This is due to the
fact that as the packet loss rate increases, the arrival
deadline for the lost packet is generally decreased because
of the less safely arrived MMT packets in the receiver
buffer.
Figure 5. Ratio of renounced retransmission for various packet loss rates
under fixed network delay of 50 ms
Figure 6 shows the saved network bandwidth resulted by
the proposed delay constrained ARQ scheme. We can
observe that the saved network bandwidth increases as the
packet loss rate increases. As the packet loss rate increases,
the ratio of the renounced retransmission also increases.
This results in avoiding unnecessary retransmission of the
out-of-date packets and therefore reduces amount of data
traffic wastefully injected into the network.
Figure 6. Saved network bandwidth for various packet loss rates under
fixed network delay of 50 ms
In the second experiment, the packet loss rate was set to
10% and the network delay has varied from 100 ms to 400
ms by the NIST-Net network emulator. When compared to
the simulation conditions for the first experiment, the
channel condition significantly deteriorated, which showed
a higher packet loss rate and much longer network delays
experienced during the video streaming.
Figure 7 shows the ratio of the number of renounced
retransmissions to the total number of retransmission
requests for various network delays under the same test
condition. It is evident that the ratio of the renounced
retransmissions significantly increased as the network
delays increased. This was due to the fact that, as the
network delay increased, the arrival deadline for the lost
packet generally decreased on account of fewer safely
arrived MMT packets in the receiver buffer.
Figure 7. Ratio of renounced retransmission for various network delays
under the fixed packet loss rate of 10%
64Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
72 / 79
Figure 8 shows the saved network bandwidth resulting
from the proposed scheme. We can observe that the saved
network bandwidth increased as the network delay increased.
As the network delay increased, the ratio of the renounced
retransmission also increased. This resulted in the avoidance
of unnecessary retransmissions of out-of-date packets and
therefore reduced a significant amount of data traffic
wastefully injected into the network. As shown in Figure 8,
approximately 190 Kbps were saved for the network delay
case of 400 ms. This saved bit-rate amount corresponded to
13% of the total data traffic generated by the transmitted
MMT packets.
Figure 8. Saved network bandwidth for various network delays under the
fixed packet loss rate of 10%
(a) packet loss rate of 5%
(b) packet loss rate of 7%
Figure 9. Comparison of RTT and arrival deadline for 100 randomly chosen
lost packets during video streaming over various packet loss rates
In Figure 9, we compare RTT and arrival deadline values
for one hundred randomly chosen lost packets during video
streaming over packet loss rates of 5% and 7%. If the RTT
is greater than the arrival deadline, the sending entity
decides not to retransmit the requested lost packet to the
receiving entity. Otherwise, the sending entity decides to
retransmit the lost packet to the receiving entity. As shown
in Figure 7, as the packet loss rate increases, the portion of
the lost packet which has greater RTT value than the arrival
deadline also increases.
V. C
ONCLUSIONS
In this paper, we proposed a delay constrained ARQ
scheme to enhance the effectiveness of the basic ARQ
process of MPEG MMT standard for video streaming over
Internet. Using the proposed ARQ scheme, it is possible to
avoid unnecessary retransmission of the out-of-date packets
and therefore minimize the probability of wastefully
retransmitted packets.
A
CKNOWLEDGMENT
This research was supported by the MSIP(Ministry of
Science, ICT and Future Planning), Korea, under the
ITRC(Information Technology Research Center) support
program (IITP-2015-H8501-15-1001) supervised by the
IITP(Institute for Information and Communications
Technology Promotion).
R
EFERENCES
[1] S. Aoki, K. Aoki, H. Hamada, Y. Kanatsugu, M. Yamamoto,
and K. Aizawa, “A new transport scheme for hybrid delivery
of content over broadcast and broadband,” Proc. Broadband
Multimedia Systems and Broadcasting (BMSB), 2011, pp. 1-6.
[2] Y. Lim, K. Park, J. Lee, S. Aoki, and G. Fernando, “MMT:
An emerging MPEG standard for multimedia delivery over
the Internet,” IEEE Multimedia, vol. 20, no. 1, Jan. 2013, pp.
80-85.
[3] ISO/IEC 23008-2, High efficiency coding and media delivery
in heterogeneous environments – MPEG-H Part 2: High
Efficiency Video Coding (HEVC), 2013.
[4] ISO/IEC 23008-1, High efficiency coding and media delivery
in heterogeneous environments – MPEG-H Part 1: MPEG
Media Transport (MMT), 2014.
[5] Y. Lim, S. Aoki, I. Bouazizi, and J. Song, “New MPEG
transport standard for next generation hybrid broadcasting
system with IP,” IEEE Trans. Broadcasting, vol. 60, no. 2,
Feb. 2014, pp. 160-169.
[6] ISO/IEC 23008-1 AMD 1, Additional technologies for MPEG
Media Transport (MMT), 2014.
[7] Y. Wang, and Q. Zhu, “Error control and concealment for
video communication: A review,” Proc. of the IEEE, vol. 86,
no. 5, May 1998, pp. 974-997.
[8] M. Podolsky, S. McCanne, and M. Vetterli, “Soft ARQ for
layered streaming media,” Journal of VLSI Signal Processing,
vol. 27, 2001, pp. 81-97.
[9] ISO/IEC JTC1/SC29/WG11 m32363, Update of ARQ
feedback message for delay-constrained retransmission in
MMT, 2014.
[10] ISO/IEC JTC1/SC29/WG11 m35228, Update of
Implementation guideline for delay constrained ARQ in
MMT, 2014.
[11] NIST-Net, Software provided by National Institute of
Standards and Technology (NIST), available on-line at
http://www-x.antd.nist.gov/nistnet, [retrieved: December,
2014].
65Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
73 / 79
On Cost-Reduced Channel Changing for Mobile IPTV Services in LTE-Advanced
Systems
Hyeungwoo Lee
Network Development Group
SAMSUNG SDS Co. Ltd.
Suwon, Gyeonggi-do, 443-822, Republic of Korea
e-mail: hw77.lee@samsung.com (zen016@naver.com)
Jae-Young Choi and Jongpil Jeong
College of Information & Communication Engineering
Sungkyunkwan University
Suwon, Gyeonggi-do, 440-746, Republic of Korea
e-mail: {jaeychoi, jpjeong}@skku.edu
Abstract— Due to the development of mobile communication
technologies, multimedia services for mobile devices or other
services have begun to be commercialized. In this paper, we
propose a method that minimizes channel switching time by
pre-computing the user’s preferred channels in the LTE-
Advanced System. The existing composition of the Multimedia
Broadcast Multicast Service (MBMS) using the Internet Group
Management Protocol (IGMP) transfers data using multicast
from the Broadcast Multicast Service Center (BM-SC) to
evolved Node B (eNB), and with broadcast from eNB to the
User Equipment (UE). Therefore, in the case of channel
switching, the IGMP Join report should be transferred to the
BM-SC, where it then waits for the user’s content to be
retransferred from eNB. However, in the case of our proposed
method, the subscriber channel technique of priorities is used
to minimize the wait times between channel switching as the
UE broadcasts its favorite channels in advance. Our
performance is evaluated using mathematical modeling and
shows that the delay time can be reduced from approximately
30 to 79% in terms of the channel switching time.
Keywords-LTE-Advanced; MBMS; MBSFN; IGMP; Mobile
IPTV; Content Database.
I. INTRODUCTION
The use of the Internet has increased greatly due to
multimedia services. It is expected to increase by at least
1000-fold by 2020 [1]. In wireless environments, as the
quality of the internet connection has improved, the demand
for multimedia services such as Internet Protocol Television
(IPTV), Video on Demand (VOD), and Voice over IP (VoIP),
using a mobile IPTV is steadily increasing. IPTV is a
communication and broadcasting convergence service based
on IP that provides a broadcast service to the user terminal,
and it can provide an interactive service according to the
user’s demands. In particular, mobile IPTV is mobility-
granted to the existing IPTV. The Multimedia Broadcast
Multicast Service (MBMS) initially appeared in Release 6.
The MBMS was introduced in order to provide a service in
Wideband Code Division Multiple Access (WCDMA). The
most important technique of the MBMS is the Multimedia
Broadcast Multicast Service Single Frequency Network
(MBSFN) transmission scheme, which has been added in
Release 7. The 3GPP has standardized tasks to provide Long-
Term Evolution (LTE) standardization and evolved MBMS
(eMBMS) in Release 8, as well as including eMBMS
standards in Release 9. In addition, the eMBMS features
(counting techniques to determine the number of user
terminals interested in a multimedia broadcast, the broadcast
service to support the mobility of the receiving user terminal
technology) have been added in Release 10/11, and are
progressing towards standardization in Release 12 [2].
The MBMS is an effective method for delivering
multimedia content to multiple destinations by allowing for
the sharing of resources more efficiently. The LTE-
Advanced system has two methods for supporting the
MBMS. One is a single-cell MBMS transmission method
used to transmit the MBMS content to users in a single
evolved Node B (eNB). Another is a multi-cell MBMS
transmission method for simultaneously transmitting MBMS
content to the users that belong to a group of eNBs. The
MBSFN transmission scheme is greatly improved in terms of
the spectral efficiency in order to improve the Signal to
Interference-Plus-Noise Rate (SINR) with compared to the
Universal Mobile Tele-communications System (UMTS) for
MBMS. This is very useful for the cell edge, which is
considered to be inter-cell interference in the UMTS, because
the interference signal is decreased and the received signal is
increased simultaneously. The MBSFN generally provides
better performance compared to single-cell Point to Point
(PTP) or Point to Multipoint (PTM) transmission [3]-[4].
As a result, the MBMS transmits content by linking one
eNB to multiple UEs. If any UE requests are connected to the
applicable eNB, even if the UE does not request the specific
channel, along with the channel requested by the other UE,
the MBMS can receive content due to the broadcasting
method. When the UE changes channels, if the UE is already
receiving the channel via broadcasting, there is no time delay
related to the channel change. If not, it can be received after
sending the desired channel to the Broadcast Multicast
Service Center (BM-SC) via the Internet Group Management
Protocol (IGMP) Join report. To support the improved
content in LTE-Advanced/eMBMS environments, we
propose a method for minimizing the channel switching time
by using channel priority assignment interlinked with the
Content Database.
This paper is organized as follows: Section 2 describes
the related work. In Section 3, we describe the improvement
of the content transmission using the channel priority scheme
with an IGMP. Section 4 shows the numerical results with
the analytical model, and finally we conclude with Section 5.
66Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
74 / 79
II. RELATED WORK
This section describes the LTE-Advanced System,
Mobile IPTV and eMBMS.
A. LTE-Advanced System
LTE (Rel. 8/9) is based on the OFDM transmission
scheme, and can be described as the evolution of 3G mobile
communication technology, such as Code Division Multiple
Access (CDMA), WCDMA and 3.5G High Speed Packet
Access (HSPA). The LTE system operates at a rate three
times higher than the 3.5G transmission system in the
Downlink and the latest standard for LTE, LTE Advanced,
seeks to achieve a rate more than twice as high as that found
in LTE [5]-[8].
LTE and LTE-Advanced technologies increase the data
transmission rate, and many technologies are being studied in
order to support seamless service. Multiple Input Multiple
Output (MIMO) technology in particular is being studied to
make high transmission rate a core technology of LTE and
LTE-Advanced systems. MIMO technology has an
advantage in that there is no help from electric power or
frequency, and it can obtain an antenna number proportionate
to the channel capacity. Increasing the performance of
mobile communication systems represents an important
avenue of research. MIMO technology has rapidly evolved
from Single-User MIMO to Multi-User MIMO, and recently
to Mass-MIMO [14].
Figure 1. Antenna techniques of LTE and LTE-Advanced.
Figure 1 shows the antenna port and the spatial
multiplexing layer in LTE (Rel. 8/9) and LTE-Advanced
(Rel. 10) (Assuming that the UE has 4 receivers – 4 x 4). The
downlink of LTE is up to support four transmit spatial layers,
and the uplink (assuming that the eNB use the diversity 1 x
2) is up to support one per UE. LTE is not supported by a
multi-antenna transmitter in order to simplify the baseline
UE. However, multi-user spatial multiplexing (Multi-User
MIMO) is supported. For MU-MIMO, the 2 UEs transmit in
the same frequency and time, and eNB has to separate the
UEs based on the spatial attributes. The gain of the uplink
capacity, including multi-user spatial multiplexing, is
generated. The maximum data transfer rate of a single user is
not improved. To improve the peak data rate of a single user,
in order to meet the requirements of the ITU-R (radio
communication sector) for spectrum efficiency, LTE-
Advanced may designate up to 8 downlink layers. This
allows for 8x8 spatial multiplexing on the downlink where
eight receivers are required for the UE. The UE has been
specified to support up to four transmitters. Therefore, when
combined with the four receivers, the eNB allows for the
transmission of a 4x4 uplink.
B. Mobile IPTV
IPTV is a the TV service provider that provides services
to the user via the IP network. In addition, the user has
received the enhanced TV service instead of the traditional
TV service through a technique called IP. Already we have
been using a similar service, such as videos and web search
using a computer or laptop via the Internet. The opportunity
is in mobile IPTV, where users cannot watch TV services as
one-way, but instead watch real-time two-way [13]. In
addition, it is possible to use the enhanced TV service over
the desired configuration. Moreover, once Korea had
obtained popular mobile TV, such as DMB, it had a
technique where you can always watch where you want. It is
a concept known as Take-Out TV. However, this model of
mobile TV technology extends the existing fixed-TV service,
which had limitations with traditional TV services, where in
particular there was still dissatisfaction with the one-way
service [2]. In order to solve this problem, techniques for
two-way service were introduced using mobile IPTV-based
IEEE802.16, 3GPP and a variety of communication
technologies (for example, 3GPP2 / BCMCS, Qualcomm /
MediaFLO, IMNA / BCAST, etc.). Figure 2 shows the
configuration of a mobile IPTV.
Figure 2. Configuration of mobile IPTV.
Mobile IPTV can support services anywhere, anytime
and also supports seamless mobility in the wireless
environment, even when moving to another radio link with a
mobility supporting technology, such as the handover
duration of the IPTV service. Especially if there are
characteristics of a different radio link, technical issues can
occur, but these technologies are expanding the research
scope of the technology because they are able to expand the
available range of mobile IPTV standardization that
organizations are required to have [14]. In addition, the
mobility studies in the Next Generation Network (NGN),
ITU-T SG13 and SG19 in charge of the mobile
67Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
75 / 79
communication network standards fall jointly under the
NGN-SGI (Global Standards Initiative) to form a group
called Mobility Management, and the standardization
progress and issues, such as Fixed Mobile Convergence
(FMC), are also being studied at the same time [5]. Recently,
mobile IPTV technology has been secured for the QoS and
QoE of the content, security, interoperability and openness,
anytime and anywhere, which aims to provide services
across any device, through the combining of the next
generation of web technology oriented user engagement and
personalization and service convergence [16].
C. eMBMS
3GPP has introduced 3G / 4G user multicast information,
as well as MBMS (Multimedia Broadcast / Multicast
Service). MBMS may share resources efficiently, as it is an
efficient method for providing multimedia content to
multiple destinations. MBMS in the "LTE" context of 3G
systems are evolving to become eMBMS [2]. LTE's eMBMS
aims to provide broadcast and multicast service combining
the high efficiency of the spectrum capacity. LTE-Advanced
is designed to share services, such as digital video / audio
broadcast.
Figure 3. eMBMS architecture.
In Figure 3, eMBMS shows the UE how to perform the
broadcast / multicast in the LTE-Advanced network [9]. It
can send a media signal simultaneously to multiple receivers
in the same area using the same eNB. In addition, the
interactive features of the eMBMS system can dynamically
interact with the broadcast network. To support the eMBMS
service without having to change the entire structure of the
existing service, the addition of a new node such as BM-SC
can be supported. The eMBMS architecture is compatible
with the SGSN and GGSN, such as 2G / GSM or the 3G
UMTS packet core node based on the packet core domain.
The eMBMS configuration of the 3GPP / LTE is shown in
Figure 3. An eMBMS gateway exists between BM-SC and
the eNB and performs functions such as MBMS packet
transfer to the eNB by IP Multicast, user plane data header
compression, and session control signaling (session start /
stop). The MCE is responsible for allocating radio resources
used by all eNBs belonging to one MBSFN area [15].
We first define a new interface SGi-mb and SG-mb for
the BM-SC and MBMS Gateway communication. The SG-
mb interface delivers authentication information and service
management to the control plane. IP packets of the
multimedia data are transmitted through the user plane
interface’s SGi-mb.
IGMP is split into two versions. It is a protocol for
managing the large group membership of a multicast router
and identifying a member of a multicast group on a network
host. IGMPv1 is described in RFC 11/12 and IGMPv2 is
described in RFC 2236. IGMPv2 is used by default in the
router [10]. The operation is performed through joining,
monitoring, member continuation, and leaving, as well as
when a multicast group joins the router and sends a General
Query message at 125-second intervals, and notifies the
router to leave (IGMPv1 does not inform). Recently IGMPv3
has been under development by RFC 3376 [11]. An IGMP
message is composed of 8 bytes, where the first four bits
discriminate the IGMP version. Consisting of 16 bits in the
checksum, there is a group consisting of a 32-bit address (IP
address of Class D). The time to query is when the group
address is set to zero, since that is when the report has a
value that represents the address of the group.
Figure 4. IGMP flow.
Figure 4 is a schematic view showing the basic operation
of the IGMP. When the first process is to participate in the
group, the host sends an IGMP Join message. The host does
not send the report when the first process and last process
leaves the group, while the router periodically sends an
IGMP Query message to check whether the host is part of a
group. The host then sends an IGMP Report message for
each group belonging to its own process, and responds to the
IGMP Query message.
Numerous standards and research papers for the Mobile
IPTV service are presented and most of them address the cost
by cell range. Our research, on the other hand, focuses on the
improvement of the quality of user experience through
minimizing the time of channel switching concerning MBMS
service in LTE/LTE-A system.
III. SERVICE CONTROL OF CONTENT DATABASE
At present, several UEs are connected to one eNB, and
the eMBMS transmits the content using a point-to-multipoint
method. Therefore, the UE receives not only the requesting
channels, but also unsolicited channels. Other UEs then
request content in the same eNB. The channel change using
IGMP is shown in Figure 5. After sending the IGMP JOIN
report from the UE to BM-SC, the channel content packet
may be transmitted from the eNB to UE.
68Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
76 / 79
Figure 5. Channel selection procedure in eMBMS.
Figure 6 describes the process for channel change. If a
UE changes from channel#1 to channel#2, the IGMP JOIN
message for channel#2, the IGMP LEAVE message for
channel#1, and transmits the content for channel#2, this is a
source of service quality degradation due to the waiting time
to perform the channel change.
Figure 6. Flow of channel selection in mobile IPTV.
As shown in Figure 6, the channels are transmitted by
multicasting from the content provider to the eNB, and then
the channels requested by the UEs are transmitted by
broadcasting from the eNB to UE. Therefore, when the UE
changes the channel for new contents, if the UE already
receives the new contents representing the requested
information by broadcast message, we assume that there is
no delay time for the channel change. However, there cannot
be an inefficient waste of broadcast packets for all channels
regardless of the subscriber. We propose a method of
broadcasting with respect to the expected channel to be used.
Figure 7 shows a representation where a database node
for the user to analyze the existing eMBMS system has been
added. When the channel change request comes from the UE,
if the channel is currently being broadcast, it sends the
content of the immediately changed channel. If the channel is
not being broadcast, it sends a message to the BM-SC
requesting the content of the requested channel. This is the
same method as the existing transmission system. Since the
BM-SC has a preference pattern for UE channels, it is
transmitted to the UE by the broadcast channel with selected
channels that are decided by the BM-SC / content database.
This method eliminates the delay time according to the
channel change of the UE. The communication between the
content provider / content database and BM-SC using this
method is performed in the system backend, which reduces
the burden of the UE and does not affect the existing channel
switching time.
Figure 7. User rating channel database additional configuration.
IV. PERFORMANCE EVALUATION
In this section, we perform a mathematical analysis to
assess the performance of MBMS channel switching time in
terms of the delay time, and present some numerical results
to evaluate the comparison of the channel change.
A. Modeling Time Cost Analysis
Table I describes the parameters which were evaluated in
the case with and without a subscriber’s favorite channels for
the channel change time.
TABLE I. PARAMETERS USED IN NUMERICAL ANALYSIS
Parameter
Description
u
U
I
IGMP messages via the Uu interface
u
U BC
I
Air channels transmitted by a broadcast message
1M SG
I
M1/SG-mb interface using IGMP message
(M1: Interface among BM-SC↔GW↔eNB)
u
U query
I
Multicast Address Query messages via the Uu interface
query
I
Multicast Address Query message between eNB and
BM-SC
x
I
Subscriber-specific favorite channel request / response
messages
2x
I
Subscriber-specific favorite channel request / response
messages
2x Sort
I
Calculated per subscriber channel preferences at the
content database
mc
I
Multicast message from the BM-SC
MBMS
I
Total time of MBMS
Hence, The MBMS time is calculated as follows;
1
2( ) 2( )
uu
MBMS air U U BC core M SG mc
I K I I K I I
69Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
77 / 79
Using improved functionality, the MBMS time can be
classified into two. The first is when it contains a
subscriber’s favorite channels when broadcasting channel
information:
_1
2( ) 2( )
uu
MBMS with CH air U U BC core M SG
I K I I K I
The second is when it does not contain a subscriber’s
favorite channels in broadcasting channel information:
_
1 2 2
2( )
2( )
uu
MBMS with CH air U U BC
core M SG x x x sort mc
I K I I
K I I I I I
air
K
is a proportion of the air interface and
core
K
is a
proportion of the core interface (
air
K
+
core
K
=1) [12]. In the
LTE system, we made calculations by setting a different
weight for the latency time. However, the actual value of the
air interface and the system performance is determined.
B. Numerical Results
Based on the analysis given so far, we compare the
performance of the existing and proposed schemes. For
numerical analysis, we configure the parameter values, as
described in Table II, in which some of the values are taken
from [12]-[16].
TABLE II. PARAMETER VALUES FOR PERFORMANCE EVALUATION
Parameter
Values
air
K
0 <
air
K
< 1, (if
air
K
+
core
K
=1)
core
K
0 <
core
K
< 1, (if
air
K
+
core
K
=1)
u
U
I
0.5
u
U BC
I
0.5
1M SG
I
2.5
u
U query
I
0.5
query
I
2
x
I
1
2x
I
1
2x Sort
I
1.5
mc
I
2.5
core
K
specifies the proportion of a real-world environment
to be larger than
air
K
. The content database node increases
and the subscriber may encounter a delay according to the
calculation for the wireless channel. In this paper, the weight
of
air
K
and
core
K
was confirmed by setting three assumptions.
If there are none of the subscriber’s favorite channels, as
shown in Figure 8, the proportion of
air
K
and
core
K
may be
increased compared to the conventional MBMS channel
switching time if the subscriber's preferred channel is
included (if it is broadcast). Accordingly, the proportion of
air
K
and
core
K
can confirm the degree of improvement from
22.7 to 30.7% compared to the existing procedures. Of
course, the delays are reduced when the proportion of
core
K
is
large.
Figure 8. Comparison of the channel change.
Figure 9. Channel popularity density [13].
According to [13], with respect to channel 96 for the
popularity investigation result of the TV viewing conditions,
0≤X≤10 is 45% as shown in Figure 9, 0≤X≤15 is 65%, and
0≤X≤20 indicates a probability of 70% for 5 or more
channels if the shared content can be sent without channel
change time.
The channel change time for the subscriber’s favorite
channel is analyzed. Figure 10 shows that it is possible to
reduce the delay time from 57.5 to 61.8% in the case of 0≤ X
≤10. Figure 11 shows that it is possible to reduce the delay
time from 76.8 to 79.2 % in the case of 0≤ X ≤20.
If the most popular 96 channels are shared in a service
node, the probability of a viewer choosing among the
channels is 0.618. Figures 10 and 11 show the cumulative
channel duplication probability of 10 independent trials when
total numbers of most popular shared channels are 10 and 20.
Notice that the newly proposed method is able to let viewers
immediately watch on the shared mobile IPTV without
channel change time using the enhanced group join and leave
process for LTE / LTE-A systems, when more than 5 channel
requests are made for most popular channels.
70Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
78 / 79
Figure 10. Comparison of channel change (0≤ X ≤10).
Figure 11. Comparison of Channel Change (0≤ X ≤20).
V. CONCLUSION
In this paper, we analyze channel change techniques and
conduct a performance evaluation for mobile IPTV in LTE-
Advanced environments. The proposed method, which is a
curtailment of channel change time, can be effective for
eliminating the delay time by broadcasting from the eNB in
advance after analyzing the channel pattern of the accessing
user in the content database. This method also reduces the
burden of the UE and doesn’t affect the existing services at
all because it is performed in the back end of the BM-SC and
content database system after the user initially connects. For
the performance evaluation, we analyze the transmission of
the content without the IGMP message using the duplication
probability of the selected channel for the shared channel. If
the computational accuracy of the subscriber priority channel
in the content database is able to be increased, we can expect
to provide a better service for MBMS and improve the
performance of MBMS.
In future work, we will seek to increase the accuracy of
the method for calculating the preferred channels of the
subscriber in the content database.
ACKNOWLEDGMENTS
This research was supported by the Next-Generation
Information Computing Development Program through the
NRF funded by the Ministry of Science, ICT & Future
Planning (NRF-2014M3C4A7030503) and Basic Science
Research Program through the NRF funded by the Ministry
of Education (NRF-2010-0024695).
REFERENCES
[1] Chih-Lin, H. Jinri, D. Ran, C. Chunfeng, J. X. Jiang, and L.
Lei, "Recent Progress on C-RAN Centralization and
Cloudification" IEEE Access, Vol. 2, 2014, pp. 1030-1039.
[2] J. S. Kim and S. H. Moon, "Evolved-MBMS: Mobile IPTV
Technologies for 3GPP LTE," Information &
Communications Magazine, Vol. 30, No. 2, 2013, pp. 66-74.
[3] R. Trestian and G. M. Muntean, "Wireless Multi-Access
Environments and Quality of Service Provisioning," IGI
Global, 2012.
[4] A. Alexiou, C. Bouras, V. Kokkinos, and G. Tsichritzis,
"Performance evaluation of LTE for MBSFN transmissions,"
Wireless Networks, Vol. 18, No. 3, 2012, pp. 227-240.
[5] B. H. Shim and B. J. Lee, "Evolution of MIMO Technology,"
The Journal of Korea Information and Communications
Society, Vol. 38A, No. 8, 2013, pp. 717-723.
[6] D. Martin-Sacristan, J. F. Monserrat, J. Cabrejas-Penuelas, D.
Calabuig, S. Garrigas, and N. Cardona, "On the way towards
fourth generation mobile: 3GPP LTE and LTE-Advanced,"
EURASIP Journal on Wireless Communications and
Networking, Vol. 2009, 2009, pp. 1-10.
[7] V. Stencel, A. Muller, and P. Frank, "LTE Advanced - A
Further Evolutionary Step for Next Generation Mobile
Networks," in proceeding of 20th International Conference on
RADIOELEKTRONIKA, 2010, pp. 1-5.
[8] B. Y. Cho and J. Y. Kim, "On the Feasibility of Dual Mode
LTE-FDD/TDD and Dual Mode WiMAX-TDD/LTE-TDD,"
The Journal of the Institute of Internet Broadcasting and
Communication, Vol 10, No 5, 2010, pp. 45-55.
[9] 3GPP, "Multimedia Broadcast/Multicast Service (MBMS);
Architecture and functional description," 3GPP TS 23.246
Release 13, 2015.
[10] H. Holbrook, B. Cain, and B. Haberman, "Using Internet
Group Management Protocol Version 3 (IGMPv3) and
Multicast Listener Discovery Protocol Version 2 (MLDv2) for
Source-Specific Multicast," IETF RFC 4604, August 2006.
[11] B. Cain, S. Deering, I. Kouvelas, B. Fenner, and A.
Thyagarajan, W. Fenner, "Internet Group Management
Protocol, Version 3," IETF RFC 3376, October 2002.
[12] A. Alexiou, C. Bouras, V. Kokkinos, and E. Rekkas,
"Evaluation of the Multicast mode of MBMS," in Proceeding
of The 18th Annual IEEE International Symposium on
Personal, Indoor and Mobile Radio Communications
(PIMRC'07), 2007, pp. 1-5.
[13] E. J. Lee, S. K. Park, and H. S. Lee, "Cross-Layer Design of
Internet Group Management Protocol for Mobile IPTV
Services in WiMAX," in Procceding of 2010 Fifth
International Conference on Systems and Networks
Communications (ICSNC), 2010, pp. 294-299.
[14] Xiangyang Wang, Tao Jiang, Yin Xu and Dong Wang, "A
downlink group pre-coding scheme for MBMS in Distributed
MIMO system," Broadband Multimedia Systems and
Broadcasting (BMSB), 2014 IEEE International Symposium
on, 2014, pp. 1-4.
[15] Baburao Damerar, Chitti Babu, "Broadcast the User Chosen
Content over LTE in a Simplified Approach Using Evolved
Multimedia Broadcast Multicast Service (eMBMS) with
Optimized MCE Scheduling Algorithm," Intelligent
Computing Applications (ICICA), 2014 International
Conference on, 2014, pp. 90-94.
[16] Junyu Lai, Guolin Sun, Gun Li, Yuankai Li, and Kaiyu Qin, "
An Active Channel Changing Delay Method to Improve IPTV
Channel Availability," Computing, Communication and
Networking Technologies (ICCCNT), 2014 International
Conference on, 2014, pp. 1-7.
71Copyright (c) IARIA, 2015. ISBN: 978-1-61208-428-2
AFIN 2015 : The Seventh International Conference on Advances in Future Internet
Powered by TCPDF (www.tcpdf.org)
79 / 79
