Content uploaded by Nikola S. Nikolov
Author content
All content in this area was uploaded by Nikola S. Nikolov on Oct 30, 2017
Content may be subject to copyright.
290
A Service Recommendation Model
for the Ubiquitous Consumer Wireless World
Haiyang Zhang1, Ivan Ganchev1,2, Nikola S. Nikolov1,3, Máirtín O’Droma1
1Telecommunications Research Centre (TRC), University of Limerick, Ireland
2Department of Computer Systems, Plovdiv University “Paisii Hilendarski”, Plovdiv, Bulgaria
3Department of Computer Science and Information Systems, University of Limerick, Ireland
E-mail: {Haiyang.Zhang; Ivan.Ganchev; Nikola.Nikolov; Mairtin.ODroma}@ul.ie
Abstract— This paper describes the general service
recommendation process matched to the telecommunication
service delivery characteristics of the Ubiquitous Consumer
Wireless World (UCWW). The goal is to provide consumers with
the 'best' service instances that match their dynamic,
contextualized and personalized requirements and expectations,
thereby aligning their usage of mobile services to the always best
connected and best served (ABC&S) paradigm. A four-tiered
architectural configuration of the UCWW service
recommendation framework is proposed along with a suitable
service recommendation model. Specific and generic smart-city
application examples are outlined. Other relevant social impact
of the proposed approach is highlighted at the conclusion of the
paper.
Keywords— Ubiquitous Consumer Wireless World (UCWW);
semantic-based recommendation; heterogeneous service network;
service recommendation model
I.
I
NTRODUCTION
The Ubiquitous Consumer Wireless World (UCWW) [1] is
a significant change to the global wireless environment,
setting out a generic consumer-centric and network-
independent techno-business model foundation for future
wireless communications. The primary change the UCWW
brings is that the users become consumers instead of
subscribers, and thus potentially are able to use the mobile
service of any service provider (SP) via the 'best' available
access network of any access network provider (ANP).
One of the key UCWW features is related to the provision
of a personalized and customized list of preferred mobile
services to consumers by taking into account their preferences
and the current network- and service context [2]. The
following are some possible smart-city related scenarios:
• Smart parking service: When a consumer in her/his car
enters a university/hospital campus or such like facility, s/he
will automatically get a recommendation for the 'best' car
parking spaces, with allocation and reservation options subject
to her/his profile preferences and campus parking policies.
The recommendation will come with enhanced functions and
information options, if required by the consumer profile, e.g.,
reservation fee payment scheme; detailed directions to that
parking space on a standard navigator app or other proprietary
app; etc. Options for provision of all or part of this service,
e.g., the key parking space reservation, can be made under
other conditions, e.g., as a ‘yes’ response to ‘reserve parking
at my work-place’ pop-up on a mobile device first thing in the
morning, even before leaving from home to go to work.
• Personal-health location-reminders: The goal is to
present the consumer with up-to-date notifications about lowest
priced consumer-prescribed drugs in drugstores/pharmacies
within the geographic location of the consumer. There would
be matching service descriptions (SDs) for apps to collect and
collate the information, e.g. as part of a cloud-based service
recommendation system, from cooperating drugstores. In the
SD for such an app, alerts or reminders may be set manually
through profile policy, when the consumer is within easy reach
of a drugstore with the lowest priced drug. There are many
consumer-oriented variations of such a service as this, leading
to many ways personal-health location-reminders might
actually work for different people. Also this service can be an
integral part to support other smart-city healthy living
applications, e.g., targeted profile-based alerts about areas of
high- and low pollen count in general or as relevant as the
targeted consumer moves around the city.
A cloud-based service recommendation system has been
proposed in [3] to support consumer requirements in scenarios
such as those described above. In this paper, we present a four-
tiered configuration architecture for this service
recommendation framework (Section III) and the use of a
novel ‘service recommendation model’ (Section IV). The
novelty of these contributions, as detailed in this paper, can be
further summarized as follows:
1) The service recommendation framework for use in the
UCWW, proposed here, models services and their related
attributes as a heterogeneous information network (HIN)
and uses meta-path-based features to generate user profiles
in order to keep them as simple as possible without losing
any valuable information.
2) A combination of a semantic service description from the
service database and implicit feedback for the
recommendation task is suggested for use.
3) A discovery of user features is proposed by using
‘betweenness centrality’[4] in the scope of the HIN.
To the best of our knowledge, we are the first to propose a
user profile generation method based on a meta path in a
HIN.
The rest of the paper is organized as follows. Section II
reviews related work. Section III presents the layered
configuration of the UCWW service recommendation
2016 IEEE 8th International Conference on Intelli
g
ent S
y
stems
978-1-5090-1354-8/16/$31.00 ©2016 IEEE
291
framework, whereas section IV describes the proposed
recommendation model. Finally, section V concludes the paper
and suggests future research directions.
II. R
ELATED
W
ORK
The service recommendation process in the UCWW is able
to automatically identify the usefulness of a mobile service,
and discover and recommend the 'best' service instance to a
particular consumer, and suggest an access to it through the
'best' ANP. It can also be viewed as the process of
personalized mobile service retrieval with constrains of
consumer preferences, current access network context, and
current service context. In this section, we review two related
research areas focused on semantic-based recommendation
systems and heterogeneous information networks.
A. Sematic-Based Recommendation Systems
In recent years, semantic recommendation systems (RSs)
with incorporated common-sense and domain-specific
knowledge have attracted considerable attention. Ontology [5]
is widely used knowledge base for incorporation with
collaborative filtering (CF) systems or content-based (CB)
systems to boost their recommendation quality. Quickstep and
Foxtrot [6] are two ontological recommendation systems that
use semantic user profiles to compute collaborative
recommendations. Mobasher, et al. [7] introduce an approach
for semantically enhanced collaborative filtering, where
structured item information is used in conjunction with user-
item ratings in order to create a hybrid similarity measure for
item comparison. Linked Open Data (LOD) [8] is another
information source to support semantic-based
recommendation systems. Authors in [9] present a knowledge-
based framework using LOD for computing recommendations
from different domains. Di Noia, et al. [10] introduce a
content-based movie recommendation system that leverages
the knowledge encoded in the LOD project, where the movie
data is exploited along different features. Among semantic
RSs, the graph-based representation of the recommendation
data within an information network has demonstrated the
effectiveness of generating accurate recommendations [11].
Entity similarity measure and ranking are hot topics among
different information network analytical approaches, and a lot
of work has been done in both fields. PageRank [12] is one of
the best-known ranking algorithms in information network
analysis, which is also successfully applied to the Internet
search. It is a link analysis algorithm that works by calculating
the number and quality of edges to an object in order to
measure a rough estimate of the importance of the object [12].
SimRank [13] is another general similarity measure defined on
networks. The rule for this approach is that two entities are
considered being similar if they are connected with similar
entities.
B. Heterogeneous Information Networks
Most existing studies are focused on homogenous
information networks, where nodes or links are being treated
as being of the same type. However, heterogeneous
information networks (HINs) represent a more general model
for real-world systems, which involves multi-type objects and
multi-type links, denoting different relations, and contains
more comprehensive relations and much richer semantic
information [14]. Moreover, in many recommendation
scenarios, the item recommendation problem exists within a
HIN environment generated from additional information
related to users and items [15]. Figure 1 illustrates a movie
recommendation problem within the scope of a HIN,
consisting of different types of entities (users, movies and
genres) and different types of relationships (social relation,
rating, genre, etc.).
Figure 1. A heterogeneous information network for movie recommendations,
with users, movies and genres as nodes and their corresponding relationships.
However, the network mining approaches mentioned
above are not suitable for HINs because they are biased to
either highly visible or highly concentrated objects [16].
Several approaches have been proposed for relevance
measurement in HINs. For instance, in [16] Sun, et al. propose
the PathSim to quantitatively measure similarity of same-type
objects in a HIN, suggesting that connected objects are similar
for different reasons by different types of paths. Shi, et al. [17]
introduce a more general path-based approach, called HeteSim,
to evaluate the relatedness of objects of the same or different
type in a similar way. Both approaches are utilized by the
service recommendation model described further in this paper.
III. UCWW
S
ERVICE
R
ECOMMENDATION
F
RAMEWORK
The UCWW service recommendation framework is
proposed to work as a platform for connecting service
providers with consumers. It allows consumers to specify their
demands for services, or alternatively, it could guess these
demands on behalf of consumers when authorized to do so. In
the end, a personalized list of 'best' service instance
recommendations is delivered to each consumer, based on
her/his context information and personal preferences. The
proposed framework, depicted in Figure 2, consists of four
tiers which are described below.
292
Figure 2. The four-tiered UCWW service recommendation framework.
• The service tier models services and their related
attributes dynamically as a heterogeneous service network
(HSN), defined in section IV.A, based on the collected
information and given network schema. The HSN is used as
both a service repository and a knowledge base. It is capable
of maintaining large sets of information and their semantics,
and is updated and indexed dynamically.
• The consumer modelling tier is used to extract profile
kernels and provide an index for high-speed data-search in the
consumer profile repository. Consumer profiles are generated
by finding profile kernels in offline process, which are the
minimal set of features describing the consumer preferences at
this tier.
• The network analysis tier is the key one in this
framework. It provides a collection of path-based relevance
measurement methods for relevant entity retrieval and a
collection of path-based algorithms for profile kernels
discovery. It is designed as an algorithm toolkit to support the
functional performance of the other three tiers.
• The recommendation tier is the most external user-facing
tier. It presents the recommendation system’s facade to the
consumers. All the queries are performed through this tier. It is
concerned with various types of contextual information as
consumer requests and leverages both pre-filtering and post-
filtering approaches [18], depending on specific contextual
attributes. It includes a temporary data store for the flow of
real-time data which will be eventually filtered and merged into
the two main data repositories (the HSN and the consumer
profile repository). The recommendation engine at this tier
provides the final personalized ranked service list to consumers
by utilizing the algorithm toolkit from the network analysis
tier.
IV. R
ECOMMENDATION MODEL
A. Service Modelling
The services in the UCWW are divided into two broad
categories – access network communication services (ANCSs)
and teleserives (TSs) [19], where the TSs are accessed by
utilizing the ANCS. The ANCS are used by consumers to find
and use the 'best' access network available in the current
location, while the TSs are more complex and are based on a
set of context parameters, classified in three groups – user-,
service-, and network context.
This paper is focused on the TS modelling only. Each
service is described by a service description (SD), which
consists of service provider information, service type, and
some related tags.
Definition 1: Information network [16]. An information
network is defined as a directed graph
(, )GVE=
with an
object type mapping function
:VA
φ
→
and a linked type
mapping function
:
E
R
ϕ
→
. Each object
vV∈
belongs to
one particular object type
()vA
φ
∈
, and each link
eE∈
belongs to a particular relation
()eR
ϕ
∈
. When the number
of object types │A│ is greater than 1 or the number of relation
types │R│ is greater than 1, the network is called
heterogeneous information network; otherwise, it is a
homogeneous information network.
Definition 2: Network schema [16]. Given a
heterogeneous information network
(, )GVE=
with the
object mapping
:VA
φ
→
and link mapping
:ER
ϕ
→
,
the network schema of G is the directed graph defined over
the object type
A
with edges for
R
, denoted as
()
,
G
SAR=
.
Definition 3: Heterogeneous service network (HSN). A
HSN is a heterogeneous information network that contains four
types of objects: service (S), service type (Type), service
provider (SP), and tag (Tag), and conforms to the network
schema shown in Figure 3a.
Note that the network schema implicitly defines the types
of arcs in a heterogeneous information network. In particular,
in the HSN there are six types of arcs, as shown in Figure 3a.
Figure 3b depicts a sample HSN.
(a) (b)
Figure 3: (a) a HSN network schema;
(b) a sample HSN conforming to (a).
According to [19], service types are classified into eight
categories: communication services, messaging services,
information services, entertainment services, education services
m-commerce services, location-based services, and other
services. Each category includes all services that perform a
similar function. Thus, service types belonging to the same
category can be linked together. Tags refer to attributes that
relate to a service. These attributes can be inferred from the
sources mentioned in section III, using the existing knowledge
base in order to get semantic relations between tags.
293
B. User Modelling
Because of different types of entities and relationships, lots
of different paths exist in the HSN. In order to describe path
types in the HSN, the meta path approach proposed in [16]
could be utilized. It defines how two entity types could be
connected via different types of relationships following certain
network schema.
Definition 4: Meta path [16]. A meta path
12
12 1
...
l
R
RR
l
PA A A
+
=⎯⎯→⎯⎯→⎯⎯→
is a path defined on a
network schema
()
,
G
SAR=
.
Each meta path defines a composite relation between the
object types in the path and represents diverse semantics. For
example, with the network schema (e.g. of movies) defined in
Figure 2, one can derive plenty of meta paths between movies.
Following each meta path, movies are linked under different
assumptions. Two sample meta paths are shown below:
Service – A –Service
Service – A – Service – B – Service,
where
,{ ,, }
A
B Type Tag ServiceProvider∈
. The first
meta path indicates that services are similar because they share
the same attributes, whereas the second one indicates a more
complex meaning of similarity.
With the implicit user feedback data, extracted at the
service tier, and the meta path defined above, we propose the
user feature selection process along meta paths. If we can
understand the semantic meaning of items that a user is
interested in, and find latent features behind these items, we
can discover the entities to constitute user profile accordingly.
Definition 5: User profile. Given a set of services
()
12
,...
N
SSSS=
that consumer-user u is interested in (used,
browsed, etc.), and a set of meta paths 12
( , ... )
M
P
PP P=
, the
profile Ku for user u is the set of nodes
,
(: )
ij ij
lij
uSSSSl
PPSS S
KppP
∈∈
=∈
⑴
where both
i
S
and
j
S
belong to S, and
ij
SS
p
is a path
between Si and Sj based on the meta path Pl, and Ku is a set of
weighted vectors.
Motivated by the definition of betweenness centrality [4]
and implicit user feedback data, we propose a meta-path-based
centrality measure to discover the most important entities for
each meta path. If we use the entities with a user’s feedback as
seeds, along each meta path, nodes with centrality value
beyond certain threshold will be selected to generate user
profiles.
Definition 6: Centrality score. Given a set of items with a
user implicit feedback
12
={e , e ...e }
n
Ι
, where
i
eitem∈
, and a
meta path 12
(, ...)
n
Pttt=
with
1n
t t item== , the centrality score
for a node n on the meta path P is:
()
{: }
{: }
(| )
ijij
ij ijij
eeee P
ee I eeeeP
n
cn P
σ
σ
∈
≠∈ ∈
=
⑵
()( )
()
{: }
{: }
2
(| ) 12
ijij
ij ijij
eeee P
ee I eeee P
n
sn P NN
σ
σ
∈
≠∈ ∈
=−−
⑶
where
{: }
ijij
eeee P
σ
∈
is the total number of paths between
i
e
and
j
e
along meta path P,
()
{: }
ijij
eeee P
n
σ
∈
is the total
number of paths between
i
e and
j
e
that pass node
n
along
meta path P, where
n
can belong to any type of node.
(| )sn P
represents the normalized centrality value for node n
where
||NI=
. Hence the centrality along meta path P is a
value between 0 and 1.
C. Service Recommendation Modelling
We propose to utilize closeness centrality [20] to build the
service recommendation model. Closeness centrality is an
important concept in network analysis, which measures how
close a node is to all other nodes in the graph. With the user
profile
12
{ , ... }
uN
Kfff=
defined in Definition 5, our
recommendation model is given as:
1
(,) (, )
iu
uj
ij
fK
n
rK s dfs
∈
−
=
⑷
where
(, )
ij
dfs
is the shortest path between the user feature
and the service entity based on the service network,
n
is the
size of user profile . The services with top-N highest score will
be selected for further context filtering processing.
The overall recommendation process can be divided into
three stages: pre-filtering, semantic recommendation, and post-
filtering. Time and companion information is used to pre-filter
the recommendation candidates. For user u, a set of services Su1
is defined considering only services available at the current
time and satisfying the user companion status. The
recommendation approach described in this subsection only
works on the services available in Su1, which results in a
recommendation candidate set Su2. Finally, localized
information is used to post-filter Su2, generating the final
recommendation list Su3.
V. C
ONCLUSION
Mobile phones are currently the most popular personal
communication devices. They have formed a new media
platform for merchants with their anytime-anywhere
accessible functionalities. However, the most important
problem for merchants is how to deliver a service to the right
mobile user in the right context efficiently and effectively. The
service recommendation system considered in this paper can
294
provide a platform to assist service providers to reach their
valuable targeted users, while at the same time offering users a
list of personalized and contextualized service instances to
choose from.
The integration of a semantic-based recommendation
system into the ubiquitous consumer wireless world (UCWW)
has the potential to create an infrastructure in which
consumers-users will have access to mobile services with a
radically improved contextualization. As a consequence, this
environment is expected to radically empower individual
consumers in their decision making and thus positively
impacting the society as a whole, and it will facilitate and
enable direct consumer – service provider relationships. Such
direct relationship types will be attractive in the provision of
‘smart city services’ in that the directness allows for more
dynamic adaptability as well as potential for user-driven
service evolution. Besides benefitting the consumers, the
UCWW opens up the opportunity for stronger competition
between service providers, therefore creating a more liberal,
more open, and fairer marketplace for existing and new
service providers. In such a marketplace, service providers can
deliver a new level of services which are both much more
specialized and reaching a much larger number of mobile
users.
The recommendation model proposed in this paper could
be employed for discovering the 'best' service instances
available for use through the 'best' access network (provider),
realizing a consumer-centric always best connected and best
served (ABC&S) experience. In the future, we will seek
further elaboration of the design, followed by implementation,
testing, and evaluation of a system prototype. The expected
main research outcomes are: realization of an efficient and
effective relevance measurement approach for the UCWW
heterogeneous service network (HSN), elaboration of an
effective graph-based feature extraction method for building
the consumer profiles, design and implementation of a cloud-
based context-aware recommendation system for the 'big data'
era.
A
CKNOWLEDGMENT
This publication has been supported by the Chinese
Scholarship Council (CSC), the Telecommunications Research
Centre (TRC), University of Limerick, Ireland, and the NPD of
the Plovdiv University, Bulgaria, under Grant No. IT15-
FMIIT-004.
R
EFERENCES
[1] M. O'Droma and I. Ganchev, "Toward a ubiquitous consumer
wireless world," Wireless Communications, IEEE, vol. 14, pp. 52-63,
2007.
[2] M. O. D. Ivan Ganchev, Nikola S. Nikolov, Zhanlin Ji, "A UCWW
Cloud-Based System for Increased Service Contextualization in
Future Wireless Networks," in 2nd International Conference on
Telecommunications and Remote Sensing, 2013, pp. 69-78.
[3] I. Ganchev, Z. Ji, and M. O. Droma, "A distributed cloud-based
service recommendation system," in 2015 International Conference
on Computing and Network Communications (CoCoNet), 2015, pp.
212-215.
[4] M. Barthélemy, "Betweenness centrality in large complex networks,"
The European Physical Journal B - Condensed Matter and Complex
Systems, vol. 38, pp. 163-168, 2004/03/01 2004.
[5] S. E. Middleton, H. Alani, and D. C. De Roure, "Exploiting synergy
between ontologies and recommender systems," arXiv preprint
cs/0204012, 2002.
[6] S. E. Middleton, D. De Roure, and N. R. Shadbolt, "Ontology-based
recommender systems," in Handbook on ontologies, ed: Springer,
2009, pp. 779-796.
[7] B. Mobasher, X. Jin, and Y. Zhou, "Semantically enhanced
collaborative filtering on the web," in Web Mining: From Web to
Semantic Web, ed: Springer, 2004, pp. 57-76.
[8] C. Bizer, T. Heath, and T. Berners-Lee, "Linked data-the story so
far," Semantic Services, Interoperability and Web Applications:
Emerging Concepts, pp. 205-227, 2009.
[9] I. Fernández-Tobías, I. Cantador, M. Kaminskas, and F. Ricci, "A
generic semantic-based framework for cross-domain
recommendation," in Proceedings of the 2nd International Workshop
on Information Heterogeneity and Fusion in Recommender Systems,
2011, pp. 25-32.
[10] T. Di Noia, R. Mirizzi, V. C. Ostuni, D. Romito, and M. Zanker,
"Linked open data to support content-based recommender systems,"
in Proceedings of the 8th International Conference on Semantic
Systems, 2012, pp. 1-8.
[11] X. Tao, Y. Li, and N. Zhong, "A knowledge-based model using
ontologies for personalized web information gathering," Web
Intelligence and Agent Systems, vol. 8, pp. 235-254, 2010.
[12] S. Brin and L. Page, "The anatomy of a large-scale hypertextual Web
search engine," Comput. Netw. ISDN Syst., vol. 30, pp. 107-117,
1998.
[13] G. Jeh and J. Widom, "SimRank: a measure of structural-context
similarity," presented at the Proceedings of the eighth ACM
SIGKDD international conference on Knowledge discovery and data
mining, Edmonton, Alberta, Canada, 2002.
[14] Y. Sun and J. Han, "Meta-path-based search and mining in
heterogeneous information networks," Tsinghua Science and
Technology, vol. 18, 2013.
[15] X. Yu, X. Ren, Y. Sun, Q. Gu, B. Sturt, U. Khandelwal, et al.,
"Personalized entity recommendation: a heterogeneous information
network approach," in Proceedings of the 7th ACM international
conference on Web search and data mining, 2014, pp. 283-292.
[16] Y. Sun, J. Han, X. Yan, P. S. Yu, and T. Wu, "Pathsim: Meta path-
based top-k similarity search in heterogeneous information
networks," VLDB’11, 2011.
[17] C. Shi, X. Kong, Y. Huang, P. S. Yu, and B. Wu, "Hetesim: A
general framework for relevance measure in heterogeneous
networks," Knowledge and Data Engineering, IEEE Transactions on,
vol. 26, pp. 2479-2492, 2014.
[18] G. Adomavicius and A. Tuzhilin, "Context-aware recommender
systems," in Recommender systems handbook, ed: Springer, 2011, pp.
217-253.
[19] P. Flynn, I. Ganchev, and M. O'Droma, "WBCs-ADA vehicle and
infrastructural support in a UCWW," in Consumer Electronics, 2006.
ISCE'06. 2006 IEEE Tenth International Symposium on, 2006, pp. 1-
6.
[20] K. Okamoto, W. Chen, and X.-Y. Li, "Ranking of closeness
centrality for large-scale social networks," in Frontiers in
Algorithmics, ed: Springer, 2008, pp. 186-195.
