Content uploaded by Alexander Felfernig
Author content
All content in this area was uploaded by Alexander Felfernig
Content may be subject to copyright.
ÖGAI Journal 25/2 1
A Short Survey of
Recommendation Technologies in Travel
and Tourism
A. Felfernig, S. Gordea, D. Jannach,
E. Teppan, M. Zanker
University Klagenfurt
Institut für Wirtschaftsinformatik und Anwendungssysteme
{felfernig, gordea, jannach, teppan, zanker}@uni-klu.ac.at
Abstract
Recommendation has a long history as a successful application area of Artificial Intelli-
gence. The demand of e-commerce platforms (e.g., amazon.com) to improve the acces-
sibility of large product- and service assortments contributed to an increased popularity of
recommendation technologies. Three basic technologies supporting the personalized
recommendation of products and services are presented in this paper. In order to take
into account the focus of this special issue, we provide a discussion of the application of
those technologies in the tourism domain (e.g., recommendation of travel destinations)
with a special focus on mobile environments.
Recommendation Technologies
The increasing size and complexity of product assortments offered by e-commerce plat-
forms requires appropriate technologies which alleviate the retrieval of products by online
customers. Different recommendation technologies have been developed to help cus-
tomers to easily find the best matching product. Those technologies have been success-
fully applied in different domains such as financial services, electronic goods, or movies.
An overview of applications exploiting recommender technologies can be found in [16].
The most widespread technology is collaborative filtering (CF), which exploits user rat-
ings of products in order to identify additional products that the active user may like as
well [6]. User-based and item-based collaborative filtering are two basic variants of this
technology. As shown in Figure 1, both variants are predicting to which extend the active
user (in this case User3) would like currently unrated items. User-based approaches to
collaborative filtering try to identify the k nearest neighbours of the active user (users
having similar tastes), and calculate a prediction of the active user’s rating for a specific
item. This rating can be defined, for example, as the weighted average of the k nearest
neighbours’ ratings [6]. In the simplified example of Figure 1, User1 is found to be the
nearest neighbour (k=1) of User3 (the active user) and his/her rating for the 4th product
(‘Conspiracy Theory’) will be taken as prediction for the rating of User3 (rate=2). In con-
trast, item-based collaborative filtering is searching for items which received similar rat-
ings from other users and were also (positively) rated by the active user. In the example
2 ÖGAI Journal 25/2
given, ‘Pretty Women' has been rated by all users. This is the most similar item to ‘Con-
spiracy Theory’ and it is assumed that User3 will have the same preference for ‘Conspir-
acy Theory’ (rate=1).
Figure 1: Content-based and Collaborative Filtering Recommendation (k=1, 1=very good; 4=bad).
Content-Based Filtering (CBF) provides recommendations for preferred product catego-
ries [5]. Let us assume, the active user has rented the ‘Pretty Women’ movie - using de-
scriptions of genre, starring and price, CBF recommends, for example, ‘Runaway Bride'.
If no product categorization is available, and the items are represented only as free text
descriptions (e.g., Netnews, books, emails etc.), alternative CBF solutions based on, for
example, information retrieval techniques are available [14], [15]. They extract a set of
keywords from textual product descriptions, compute the users preferences expressed in
terms of keywords which are contained in products bought by the user, and build the list
of recommendations by searching for products that match the user’s preferences.
CF and CBF technologies exploit user preferences and allow acceptable recommenda-
tion accuracy for frequently bought products such as music or video DVDs, books, Net-
news, Internet radio etc. Amazon.com is probably the most popular online shop that im-
plements these approaches. When accessing product descriptions, a list of CF recom-
mendations is available for users in the section 'Customers who bought this item also
bought', while the content-based suggestions can be accessed via 'Look for related items
by keyword' and 'Look for similar items by category' links.
There are other types of products (in many cases high-involvement products) which are
less frequently bought and their purchase is related to higher risks (e.g., financial ser-
vices, cars, electronic goods, services in the tourism domain). When recommending such
products, recommender applications must support a more detailed elicitation of user re-
quirements. Deep domain knowledge has to be exploited in order to be able to make
more precise and more trusted recommendations. Knowledge-based (KB) recommender
technologies [4], [8] support sales processes of high involvement products. KB recom-
menders are based on a detailed description of the product domain in the form of struc-
tured product descriptions and constraints. The identification/construction of user prefer-
ences usually takes place in the context of an explicit sales dialog. The major advantage
of this type of recommendation technology is the explicit representation of product, mar-
keting and sales knowledge. Such a representation allows the calculation of explanations
which provide, for example, a detailed argumentation as to why a certain product fits to
the wishes and needs of a given customer.
Each of the presented recommendation paradigms has its own advantages and disad-
vantages. Hybrid recommenders combine two or more of these paradigms in order to
mutually eliminate disadvantages and to improve recommendation accuracy, robustness
and trust in calculated recommendations (see, e.g., [6]).
ÖGAI Journal 25/2 3
4 ÖGAI Journal 25/2
Recommender Systems in Travel and Tourism
The travel and tourism industry is one of the most important and dynamic sectors in
Business-to-consumer (B2C) e-Commerce. According to [25], already in 2003 this single
sector made up more than fifty percent of the global B2C turnover. A variety of recent
studies (e.g., by the European Travel Commission – ETC) revealed that at least in devel-
oped countries, the Web is nowadays already the primary source of information for peo-
ple when searching or booking suitable travel destinations [12]. Consequently, the do-
main has always been at the forefront of Information Technology [25] and still is a highly
attractive research area as lots of potentials are not yet fully exploited. In this context,
recommender applications can be valuable tools supporting, for example, information
search, decision making, and package assembly.
When looking at today's e-Tourism web sites we can observe that only some of the exist-
ing systems provide services that go beyond a pure booking system's functionality. Popu-
lar online 'travel agencies' like Expedia (www.expedia.com) at least aim at exploiting the
potential of Web communities by letting their customers rate individual hotels or destina-
tions. Still, in these applications the average ratings of other customers merely serve as
another piece of information for a certain hotel or destination but there is typically no rec-
ommendation service available.
The reasons why established recommendation techniques like amazon.com cannot be
directly applied to the tourism domain are manifold. Collaborative filtering techniques
work best when there exists a broad user community and each user has already rated a
significant number of items. As individual travel planning activities are typically much less
frequent like, for example, book purchases, and in addition the items themselves may
have a far more complex structure, it is hard to establish reasonable user profiles. There-
fore, many approaches aim at eliciting the preferences and requirements in a conversa-
tional dialog (e.g., [12], [19], [21]) using, for example, knowledge-based approaches [4],
[8] for generating recommendations. Online users may be different with respect to their
background knowledge, their mental models ([17], [27]), or their capabilities of expressing
their needs and requirements. Dialog design, usability aspects, and adaptivity are thus
central in application and user interface design [20]. In [21] for instance, a critique-based
dialog style is proposed which has already been successfully employed in other domains;
[12] describes a tourism advisory application based on knowledge-based personalization
and multi-step, adaptive dialogs.
The problem of 'group recommendation' is another typical aspect in tourism-related re-
commender systems, i.e., the problem of generating proposals that 'maximize' the overall
acceptance of members of a travel group that have different interests. Although this prob-
lem is not new in recommender systems (think of TV program or movie recommenders),
there is only little research in the specific context of recommender systems in tourism
(see, e.g., [3]).
Finally, another important facet which makes recommendation in the tourism domain
more complex is the fact that a single trip arrangement may consist of several, independ-
ently configurable services [20]. Typically, only pre-defined packages like 'flight and hotel'
or 'all-inclusive' arrangements are available online. As the segment of individualized
travel arrangements is constantly growing, it will be increasingly important that future
ÖGAI Journal 25/2 5
systems support such packaging services. Nevertheless, only first attempts in that direc-
tion can be found in literature today (see [3], [9], or [19]).
Mobile Systems and Recommendation
With respect to mobile recommender applications the tourism domain is a very active
area. The idea of providing context-aware information services to tourists has already a
considerable tradition. For instance [1] present their mobile tour guide that displays points
of interest (POIs) on an interactive map dated back in 1997. Since then much more ex-
amples of mobile tour guides und context-aware applications have been reported, such
as an electronic guide for the city of Lancaster [7], the COMPASS project in the Nether-
lands [24], MobiDenk - a location-aware information system for historic sites [13] or Ber-
lintainment - an entertainment guide for the city of Berlin [26].
Although a variety of mobile and context-aware applications already exists, there are still
major shortcomings. Most systems are research prototype applications that are only
evaluated in small field trials with a limited scope for usage. Many times a wider produc-
tive use is impossible for the two following reasons. First, mobile guides might have re-
strictive hardware requirements like a specific type of PDA (Portable Digital Assistant),
the availability of GPS (Global Positioning System) functionality or client-side software
installations. However, new generations of mobile phones having larger display sizes,
more standardized browsing capabilities and broadband data transfer will ease these
hardware requirements. Second, the availability of extensive and accurate resource data
is another bottleneck. For instance, a mobile restaurant recommender requires not only
the positioning coordinates of all restaurants within a specific region but also some addi-
tional qualitative data, such as the type of food served, the atmosphere perceived by
guests or the opening hours. As acquisition and maintenance of product data are quite
cost-intensive, only widespread use and acceptance of mobile recommendation applica-
tion by end-users will make the data effort worth.
Context-awareness is a common characteristic for all these systems [10], [18]. Shilit et al.
[23] name the most important aspects: 'where you are', 'who you are with' and 'what re-
sources are nearby'. Exploiting the current location of the user, her/his companions as
well as the availability of resources in her/his surrounding can increase considerably the
perceived usefulness of a mobile application. While [22] give a coherent overview of dif-
ferent levels of context-awareness implemented by mobile tourism guides, we want to
focus on the implications of context-awareness for recommendation technology.
Currently, most reported systems filter the presented information content according to
users’ current location and their additional preferences (e.g., ‘display only objects from a
specific category’). This constitutes already a considerable degree of personalization and
reduces information overload. However, such approaches do not employ traditional rec-
ommendation techniques such as content-based or collaborative filtering. Adomavicius et
al. [2] therefore developed a multidimensional approach that allows them to incorporate
contextual information with filtering applications. They understand the term contextual
information in a general way such that it encompasses any additional data dimension.
They extend the traditional two-dimensional (user x product) representation of rating data
to a n-dimensional data cube. In their experimental evaluation in the movie domain they
for instance employed the place where the movie was watched (home vs. theatre), the
time (weekday or weekend), the type of friends who were with as well as release informa-
tion on the movie indicating its novelty as contextual data dimensions.
6 ÖGAI Journal 25/2
Considering this contextual information aggravates the cold-start problems mentioned in
the introduction due to the high degree of data sparsity. Adomavicius et al. [2] therefore
introduce reduction-based estimations that outperform traditional two-dimensional re-
commender systems in their experimental setup. Another more advanced recommenda-
tion technique that has also been fielded for some experimental evaluation was pre-
sented by [21]. They developed a location-aware critiquing system that allowed its users
to determine nearby restaurants that conformed to their interactively entered criteria and
critiques on previous proposals. It is implemented by a rich-client application that has to
be installed on the PDA and communicates with the central server.
The etPlanner system [11] is currently under development by the Austrian network for e-
tourism. It focuses on widespread and actual use among tourists and therefore avoids
client-side installation requirements. One of its novelties are the support of two types of
communication paradigms with its users. First, information seekers have personalized
browsing access to categories like events, sights, restaurants or accommodations. In a
second step, users may also receive personalized push messages that inform them
about changing weather conditions if they are out hiking or make them propositions on
leisure activities based on their preferences. A first version has already been deployed for
public use. Additional recommendation features will be added for future versions when
more information-rich user profiles will be available.
Conclusion
This paper provides an overview of recommendation technologies applied in existing
commercial environments. Application examples are mainly given from the tourism do-
main where recommendation technologies hold an extremely important role. Recommen-
dation technologies in this application domain will be even more important in the future. A
special research focus has to be set on the development of recommendation formalisms
taking into account the current context (dimensions such as time, space, mood or social
environment) of the user/customer.
References
[1] G. Abowd, C. Atkeson, J. Hong, S. Long, R. Kooper and M. Pinkerton: Cyberguide: a mobile context-aware
tour guide, ACM Wireless Networks, 5(3):421-433, 1997.
[2] G. Adomavicius, R. Sankaranarayanan, S. Sen and A. Tuzhilin: Incorporating Contextual Information in
Recommender Systems Using a Multidimensional Approach, ACM Transactions on Information Systems,
23(1):103-145, 2005.
[3] L. Ardissono, A. Goy, G. Petrone, M. Segnan and P. Torasso: INTRIGUE: personalized recommendation of
tourist attractions for desktop and handset devices, Applied AI, Special Issue on Artificial Intelligence for
Cultural Heritage and Digital Libraries. 17(8-9):687-714. Taylor and Francis, 2003.
[4] Robin Burke. Knowledge-based recommender systems. Encyclopedia of Library and Information Systems,
69, 2000.
[5] D. Billsus and M. J. Pazzani. User modeling for adaptive news access. User Modeling and User-Adapted
Interaction, 10(2-3):147–180, 2000.
[6] Robin Burke. Hybrid recommender systems: Survey and experiments. User Modeling and User-Adapted
Interaction, 12(4):331–370, 2002.
[7] K. Cheverst, N. Davies, K. Michell, A. Friday, C. Efstratiou: Developing a Context-aware Electronic Tourist
Guide: Some Issues and Experiences. CHI Letters, 2(1):17-24, 2000.
[8] A. Felfernig, G. Friedrich, D. Jannach, and M. Zanker, An Environment for the Development of Knowledge-
based Recommender Applications, to appear in International Journal of Electronic Commerce (IJEC),
2007.
ÖGAI Journal 25/2 7
[9] D. R. Fesenmaier, F. Ricci, E. Schaumlechner, K. Wöber, and C. Zanellai, DIETORECS: Travel Advisory
for Multiple Decision Styles, Enter 2003, 2003.
[10] P. Hertzog, M. Torrens: Context-aware Mobile Assistants for Optimal Interaction: a Prototype for Support-
ing the Business Traveller, 9th International Conference on Intelligent User Interfaces, pp.256-258, 2004.
[11] W. Höpken, M. Fuchs and M. Zanker: etPlanner: A hybrid recommender system for mobile travel planning,
OEGAI Journal, 24(4), pp. 26-31, 2005.
[12] D. Jannach, M. Zanker, M. Jessenitschnig, O. Seidler: Developing a conversational travel advisor with
Advisor Suite. To appear in ENTER 2007, 2007.
[13] J. Krösche, J. Baldzer, S. Boll: MobiDENK-Mobile Multimedia in Monument Conservation. In: IEEE Multi-
Media, 11(2):72–77, 2004.
[14] D. Mladenic. Text-learning and related intelligent agents. IEEE Intelligent Systems, 14:44–54, 1999.
[15] R. J. Mooney, L. Roy, Content-based book recommending using learning for text categorization, Proceed-
ings of the fifth ACM conference on Digital libraries Publisher, ACM Press, pp. 195-204, 2000.
[16] M. Montaner, B. Lopez, and J. De la Rose, A Taxonomy of Recommender Agents on the Internet, Artificial
IntelligenceReview, 19:285–330, (2003).
[17] B. Pan, and D. R. Fesenmaier, Exploring the structure of travel planning on the Internet, to appear in An-
nals of Tourism Research.
[18] A. Pashtan, R. Blattler, A. Heuser and P. Scheuermann: CATIS: A Context-Aware Tourist Information
System. 4th International Workshop of Mobile Computing, 2003.
[19] F. Ricci, and F. Del Missier: Supporting Travel Decision Making through Personalized Recommendation. In
C-M Karat, J. Blom, and J. Karat (eds.), Designing Personalized User Experiences for eCommerce, Kluwer
Academic Publisher, 221-251, 2004.
[20] F. Ricci, Travel recommender Systems, IEEE Intelligent Systems, November/December 2002, pp. 55-57.
[21] F. Ricci and Q. N. Nguyen, Critique-Based Mobile Recommender Systems, OEGAI Journal, 24(4):2005.
[22] W. Schwinger, C. Grün, B. Pröll, W. Retschitzegger, A. Schauerhuber: Context-awareness in Mobile Tour-
ism Guides, Technical report – Institute of Bioinformatics JKU Linz, [ftp://ftp.ifs.uni-
linz.ac.at/pub/publications/2005/0405.pdf], 2005.
[23] B. Shilit, N. Adams and R. Want: Context-Aware Computing Applications. In: IEEE Workshop on Mobile
Computing Systems and Applications. Santa Cruz, CA, 1994.
[24] M. van Setten, S. Pokraev and J. Koolwaaij: Context-Aware Recommendations in the Mobile Tourist Appli-
cation COMPASS. International Adaptive Hypermedia Conf., LNCS 3137, pp. 235–244, 2004.
[25] H. Werthner, Intelligent Systems in Travel and Tourism, Proceedings of the 18th International Joint Confer-
ence on AI (IJCAI), Acapulco, Mexico, August 9-15, 2003.
[26] J. Wohltorf, R. Cissée and A. Rieger: BerlinTainment: An Agent-Based Context-Aware Entertainment
Planning System. IEEE Communications, 43(6):102-109, 2005.
[27] Z., Xiang and D. Fesenmaier, An analysis of two search engine interface metaphors for trip planning,
Information Technology & Tourism, 7(2):103-117, 2005.
