=Paper=
{{Paper
|id=Vol-1906/paper6
|storemode=property
|title=Co-Staying: A Social Network for Increasing the Trustworthiness of Hotel Recommendations
|pdfUrl=https://ceur-ws.org/Vol-1906/paper6.pdf
|volume=Vol-1906
|authors=Catalin-Mihai Barbu,Jürgen Ziegler
|dblpUrl=https://dblp.org/rec/conf/recsys/Barbu017
}}
==Co-Staying: A Social Network for Increasing the Trustworthiness of Hotel Recommendations==
https://ceur-ws.org/Vol-1906/paper6.pdf
Co-Staying: A Social Network for Increasing the
Trustworthiness of Hotel Recommendations
Catalin-Mihai Barbu Jürgen Ziegler
University of Duisburg-Essen University of Duisburg-Essen
Duisburg, Germany Duisburg, Germany
catalin.barbu@uni-due.de juergen.ziegler@uni-due.de
ABSTRACT the tourism domain, consider a hotel recommender that
Recommender systems attempt to match users’ preferences models the relationships that appear between users, hotels,
with items. To achieve this, they typically store and process and hotel amenities. Furthermore, the internal representation
a large amount of user profiles, item attributes, as well as an of such networks in usually not visible to users. This can
ever-increasing volume of user-generated feedback about happen for various reasons, ranging from privacy concerns
those items. By mining user-generated data, such as reviews, to the inherent difficulty of creating a meaningful illustration
a complex network consisting of users, items, and item of the structure of complex networks.
properties can be created. Exploiting this network could
allow a recommender system to identify, with greater The purpose of this short paper is to advance the state of the
accuracy, items that users are likely to find attractive based art in two ways. First, we propose a concept of a multimode
on the attributes mentioned in their past reviews as well as network for representing users, hotels, and hotel properties.
in those left by similar users. At the same time, allowing We argue that a hotel recommender can exploit multimode
users to visualize and explore the network could lead to networks to generate more suitable suggestions. Second, we
novel ways of interacting with recommender systems and examine potential interactive mechanisms that could be
might play a role in increasing the trustworthiness of developed to facilitate the presentation of the network to
recommendations. We report on a conceptual model for a users. We believe that having access to additional means of
multimode network for hotel recommendations and discuss visualizing hotel information would allow users to interact
potential interactive mechanisms that might be employed for in novel ways with the RS. Furthermore, such novel
visualizing it. interaction techniques could play a role in increasing the
trustworthiness of recommendations.
Author Keywords
Recommender systems; tourism; personalization; trust; In the following sections, we review related work on the
multimode networks; trustworthiness usage of multimode networks in RS. We also investigate
some of the typical visualization techniques that have been
ACM Classification Keywords developed so far. Next, we present our conceptual model for
H.5.2 [Information Interfaces and Presentation]: User a user-item-attribute network using references and examples
Interfaces—evaluation/methodology, graphical user from the hotel booking domain. Finally, we discuss how
interfaces (GUI), user-centered. allowing users to interact with such a network might
INTRODUCTION introduce novel ways of interacting with RS. Additionally,
Recommender systems (RS) typically build and maintain we open the discussion on whether exploiting this network
network representations of the data they store about their could play a role in increasing the trustworthiness of
users and product catalogs. One of the earliest and most well- recommendations.
known examples is the “item-to-item” model introduced by RELATED WORK
Amazon in its RS, which enabled the company to show, The emergence of web communities and the sustained
using collaborative filtering techniques, what other products growth of the user-generated content available online
were purchased by users who bought a certain product [8]. provides important opportunities for RS. Incorporating
Linking user preferences with items and subsequently social network information has the potential to alleviate
modeling the various relationships that arise between them cold-start and sparsity problems, which are inherent in
can increase the accuracy of recommendations. Most typical collaborative filtering approaches [18]. In the field of
commonly, RS employ 1-mode networks, meaning that all RS for tourism (and, more generally, for commerce), a major
nodes are of the same type (e.g., users). Less frequent is the aspect of exploiting user-generated content is the extraction
use of 2-mode networks [2], in which relationships are of topics and sentiments from reviews, e.g., to create richer
shaped between two different types of nodes, e.g., users and user models [12]. Combining social networks analysis
items. A relatively unexplored area of research concerns the techniques, such as community detection and visualizing
usage of multimode (or n-mode) networks, in which vertices techniques, with RS is, therefore, a worthwhile direction for
can be of three or even more types. To offer an example from continued research [10]. A survey of current state-of-the-art
RecTour 2017, August 27th, 2017, Como, Italy. 35 Copyright held by the author(s).
Figure 1 – Example of how a simple multimode network (right) containing three types of vertices—users (circle), topics (hexagon),
and hotels (rectangle)—could be generated from crawled user-generate content (left). In our conceptual model, users and hotels
are connected through topics, instead of directly.
methods in both fields as well as existing challenges are relevant in the context of hotel recommending, where the
highlighted, for example, in [18]. risk associated with poor choices is often higher. Visualizing
the underlying factors used to generate recommendations
Exploiting 2-mode networks for improving the quality of
could, for example, allow developers to embed trust cues
recommendations has already been achieved in various
(i.e. interface elements that allow the user to determine the
domains. One such approach was used to derive similarity
reliability of the presented information) into the presentation
information between artists and songs by exploiting data
layer [13].
collected from a music file-sharing network [14]. The
information was then used to build a 2-mode graph of users CO-STAYING NETWORK
and songs, which was subsequently used to recommend new Based on our review of the literature, there appears to be a
artists. Citation networks are also a frequent focus of research gap with respect to the usage of multimode
research [9]. The usage of multimode networks for networks for RS. At the same time, the goal of providing
recommendations is less frequent [17]. Linking users to interactive mechanisms for allowing users to explore the
items via tags to improve recommendations is of definite underlying graph has been studied less frequently.
interest in the RS community [15]. Still, tripartite graphs Example Domain and Dataset
containing users, items, and tags have been studied more in Hotel booking is a domain in which the trustworthiness of
the context of information retrieval [3]. Alternate recommendations is especially important. More generally,
approaches, for example using multiple 1-mode networks (as while selecting the domain we considered three aspects: 1)
opposed to multimode networks) for generating article The choice should carry a substantial amount of risk for the
recommendations, have also been proposed [20]. user; 2) the items should have a reasonable set of attributes
The problem of visualizing complex networks employed in that need to be considered; and 3) there should be a large
RS is also receiving increasing attention. Previous work has body of user-generated content available, in the form of
been done, for instance, on facilitating the exploration of reviews, photos, tags, and ratings, that can be leveraged in
recommended items by combining different entities (users, the presentation. Because of the first criterion, we decided
tags, and agents) [16]. Although there is ongoing research on against using the more common domain of movie
visualization techniques for tripartite networks [7], it appears recommendations. Hotel booking, on the other hand, fulfils
that such methods are usually not applied in a systematic all three conditions.
way in the context of RS. We argue that multimode networks We crawled metadata and overall 838,780 user reviews for
should not be used solely for enhancing the generation of 11,544 hotels located in five major European cities from
recommendations. Providing means to visualize the network Booking.com1. This real-world dataset ensures access to a
as well as mechanisms for interacting with it could increase representative and interesting subset of hotels, covering as
the transparency and control of the RS [16]. The lack of many types of amenities as possible. Furthermore, it features
transparency exhibited by many modern RS is frequently a diverse set of reviews contributed by various types of
cited as having a detrimental effect on the perceived travelers—and who are travelling for different purposes—
trustworthiness of such systems [6]. Trust is especially
1
http://www.booking.com/
RecTour 2017, August 27th, 2017, Como, Italy. 36 Copyright held by the author(s).
thereby maximizing the variety of user opinions and
arguments. A characteristic of Booking.com is that all
reviews on its site are verified, meaning they are written by
people who have stayed in those hotels. This feature reduces
the number of fake reviews.
Conceptual Model
Our conceptual model, to which we will refer in the
following sections as a “co-staying network”, has three types
of vertices: users, topics (i.e. hotel attributes), and hotels. In
contrast to a typical 2-mode network, users are not linked
directly to the hotels in which they stayed. Instead, an edge
is first created between the user and a topic. A second edge Figure 2 – Example of a soft link from a topic to a hotel that is
then links the topic to the hotel for which the review was part of the same chain as the hotel for which the review was
submitted (Figure 1). This process is repeated iteratively for written. Soft links can only be created for specific topics.
each topic in a review as well as for all user-contributed
reviews. If a review contains references to more than one This approach lets the RS exploit user opinions about
topic (as is often the case in hotel reviews), there will be selected topics when suggesting recommendations that
several paths between its author and the hotel, each passing might otherwise not have enough reviews. Among the topics
through a topic. Furthermore, if a user mentioned the same that could be shared in this way are those concerning room
topic in more than one review, a separate pair of edges will layout and furnishings, breakfast, and hotel facilities. Topics
be created for each instance. such as those related to the hotel’s location or the service
quality, on the other hand, would not be shared between
The crawled user reviews in the dataset are mined to extract hotels (Figure 2).
topics. First, we identify attribute-value pairs such as
“comfortable bed” in the reviews. Then, these pairs are Clusters can be identified for each type of vertex in the co-
merged with others that have the same meaning, e.g., “comfy staying network. Topics can be characterized by their
bed”. Next, using sentiment analysis, pairs are classified as overarching category (i.e. hotel, room, or service).
positive or negative hotel properties. Finally, pairs that Furthermore, by analyzing travel and review patterns,
describe properties related to, for example, a hotel room, are additional relationships between hotel topics could be
classified and clustered together. Broadly, topics are divided identified. Hotels can be clustered based on whether they are
into three main categories: room attributes (e.g., bed, part of a chain as well as by looking at the most common
shower, minibar), hotel attributes (e.g., location, swimming topics that characterize them. Finally, user similarity
pool, parking), and hotel services (e.g., breakfast, Wi-Fi measures can be extended to include, in addition to
quality, friendliness of staff). The complete procedure is demographic data (e.g., age range, country of origin) and
described in detail in [4]. Topic disambiguation techniques rating behavior, also experience (e.g., based on the number
are employed to prevent duplication and correct spelling of contributed reviews, frequency of travel, and the types of
errors. For example, the terms “wi-fi”, “wifi”, “wireless hotels visited). Preliminary work on these aspects has
network”, “wireless internet”, or “wifus” (an incorrect already been reported in [1].
spelling that is encountered relatively often in user reviews) The proposed model aims to improve upon traditional
should all be filed under a single attribute, e.g., “Wi-Fi”. approaches to hotel recommending by creating stronger
The user’s sentiment regarding each topic is classified with links between users, hotels, and hotel attributes. This would
respect to its polarity (i.e. positive, neutral, or negative) and allow users to explore, in addition to the details of the
strength. These two attributes are normalized and encoded recommended items, the public profiles and preferences of
in the edges as a single value in the interval [-1,1]. Values those guests whose reviews were exploited for generating
closer to the left side of the interval are indicative of strong the recommendations. Various forms of interactive
negative sentiments about a topic (e.g., “terrible breakfast”). mechanisms could be developed to facilitate this kind of
Similarly, higher positive values are associated with topics exploration.
that a user has praised in a review (e.g., “wonderful Interactive Mechanisms
location”). The initial focal point of the network would be the hotel that
is being recommended. Users might ask themselves, “What
If a hotel is part of a chain, our model also allows the creation
do people talk about when reviewing this hotel?” The most
of additional “soft links” from a topic to the rest of the hotels
common topics (ideally tailored to fit the user’s interests)
in that chain. This is based on the premise that these hotels
could be shown radiating from the hotel. These should be
share many characteristics—though not all. Thus, it is likely
clustered by category. Edges would connect these topics to
(and typically advertised as such by hotel brands) that
the most representative users. Ensuring that the co-staying
travelers would have comparable experiences and access to
network remains accessible to users of the RS is a non-trivial
similar amenities in every location that is part of a franchise.
task. One mechanism could involve providing sufficient
RecTour 2017, August 27th, 2017, Como, Italy. 37 Copyright held by the author(s).
criteria to filter topics and users to avoid information presented. Moreover, the proposed interactive mechanisms
overload. Furthermore, the proportion of the co-staying are meant to increase user control over the presentation of
network that is visible to the user should also be controlled, the recommended items. Thus, the trustworthiness of the
for example by implementing a “zoom in / zoom out” design recommendation should increase [6]. Developing evaluation
pattern. criteria for measuring the effects of these novel interactions
on the trustworthiness of the recommendations is planned for
Users should also be able to refocus the network based on
future work.
their interests or goals. The system should allow seamless
transition between vertices, for example between hotels, Our approach should also alleviate, to some extent, the data
from a hotel to a topic, or from a topic to the users who sparsity problem by sharing topics—and, therefore, user
referenced it in their reviews. Interacting with a topic should opinions—about hotels that are part of a chain. However,
bring up the review snippets in which it is mentioned. The this will continue to remain an issue in the case of isolated
user could then expand reviews that look promising to read hotel vertices, for which not enough user-generated
them fully; alternatively, less interesting snippets could be information is available from the network. This aspect will
hidden completely. be investigated in future work.
Trust in online reviews has been shown to depend on the ACKNOWLEDGMENTS
credibility of the source [19]. Thus, users should be afforded This work is supported by the German Research Foundation
the possibility to explore the public profiles of reviewers that (DFG) under grant No. GRK 2167, Research Training Group
have contributed opinions about topics of interest. Public "User-Centred Social Media".
profiles might contain information about contributed REFERENCES
reviews, visited hotels, frequently-mentioned topics, as well 1. Barbu, C.-M. & Ziegler, J. User Model Dimensions for
as any demographic data that users may want to share (e.g., Personalizing the Presentation of Recommendations.
purpose of travel, typical number of nights spent in hotels, In Proc. IntRS '17. ACM (to appear).
average price paid). Trustworthiness cues embedded in the
personal profiles could help users decide whether to 2. Borgatti, S. P. & Everett, M. G. Network analysis of 2-
consider—or not—the comments left by other users in the mode data. Social networks 19, 243–269 (1997).
network regarding a certain recommendation. We expect 3. Clements, M., de Vries, A. P., & Reinders, M. J.
that this might lead to a better calibration between the user’s Optimizing single term queries using a personalized
trust in the RS and the system's actual trustworthiness [5]. Markov random walk over the social graph. In ECIR
DISCUSSION AND FUTURE WORK
Workshop on Exploiting Semantic Annotations in
Including hotel topics as an additional type of vertex in the Information Retrieval (ESAIR '08), Springer (2008).
network is expected to allow the RS to identify, with greater 4. Feuerbach, J., Loepp, B., Barbu, C.-M., & Ziegler, J.
accuracy, which items users are likely to find attractive Enhancing an Interactive Recommendation System
based on the attributes mentioned in their reviews as well as with Review-based Information Filtering. In Proc.
in reviews left by similar users [3,11]. At the same time, the IntRS '17. ACM (to appear).
system could extract and present, for each recommended 5. Harman, J. L., O’Donovan, J., Abdelzaher, T., &
item, the experiences of other people who are interested in Gonzalez, C. Dynamics of human trust in
the same combination of topics as the current user. The co- recommender systems. In Proc. RecSys '14, 305–308,
staying network might further facilitate the discovery of ACM (2014).
users with similar preferences and who have already stayed
in the same hotel or in hotels with similar characteristics to 6. Konstan, J. A. & Riedl, J. Recommender systems:
the one that a person is considering. We believe that this From algorithms to user experience. User Mod. and
could thus introduce novel ways of interacting with RS. For User-Adapted Interaction 22, 1-2, 101–123 (2012).
example, someone who has a strong preference for soft beds 7. Lambiotte, R. & and Ausloos, M. Collaborative
would be able to explore the opinions of other travelers who tagging as a tripartite network. In Proc. ICCS '06,
share this preference. Taking this a step further, the user 1114–1117, Springer-Verlag Berlin Heidelberg (2006).
might then ask herself, “What other preferences do such 8. Linden, G., Smith, B., & York, J. Amazon. com
people have?” Personalizing the presentation to facilitate the recommendations: Item-to-item collaborative filtering.
exploration of similar people’s expectations could allow IEEE Internet computing, 7(1), 76–80. IEEE (2003).
users to discover new preferences that they would not have
considered otherwise. A prototype implementation of the 9. Luong, N. T., Nguyen, T. T., Hwang, D., Lee, C. H., &
presented co-staying concept and interactive mechanisms is Jung, J. J. Similarity-based Complex Publication
currently under development. Network Analytics for Recommending Potential
Collaborations. J.UCS, 21(6), 871–889 (2015).
We believe that our co-staying network concept has the
10. Maivizhi, R., Sendhilkumar, S., & Mahalakshmi, G. S.
potential to increase the transparency of the RS by providing
A Survey of Tools for Community Detection and
visual clues as to why certain recommendations are
RecTour 2017, August 27th, 2017, Como, Italy. 38 Copyright held by the author(s).
Mining in Social Networks. In Proc. ICIA '16, 71–79, 16. Verbert, K., Parra, D., Brusilovsky, P., & Duval, E.
ACM (2016). Visualizing recommendations to support exploration,
11. McAuley, J. & Leskovec, J. Hidden factors and hidden transparency and controllability. In Proc. IUI '13, 351–
topics: Understanding rating dimensions with review 362, ACM (2013).
text. In Proc. RecSys '13, 165–172, ACM (2013). 17. Wasserman, S. & Faust, K. Social Network Analysis:
12. Rossetti, M., Stella, F., & Zanker, M. Analyzing user Methods and Applications. Cambridge Univ. Press
reviews in tourism with topic models. Information (1994).
Technology & Tourism, 16(1), 5–21 (2016). 18. Xu, G., Wu, Z., Zhang, Y., & Cao, J. Social
13. Sacha, D., Senaratne, H., Kwon, B. C., Ellis, G., & networking meets recommender systems: survey.
Keim, D. A. The role of uncertainty, awareness, and International Journal of Social Network Mining, 2(1),
trust in visual analytics. IEEE Trans. Vis. Comp. 64–100 (2015).
Graph., 22-1, 240–249 (2016). 19. Yoo, K. H., Lee, Y., Gretzel, U., & Fesenmaier, D. R.
Trust in travel-related consumer generated media.
14. Shavitt, Y. & Weinsberg, U. Song clustering using
Information and communication technologies in
peer-to-peer co-occurrences. In Proc. ISM ‘09. 11th
tourism 2009, 49–59, Springer (2009).
IEEE International Symposium on Multimedia, 471–
476, IEEE (2009). 20. Zhou, D., Zhu, S., Yu, K., Song, X., Tseng, B. L., Zha,
H., & Giles, C. L. Learning multiple graphs for
15. Sen, S., Vig, J., & Riedl, J. Tagommenders: connecting
document recommendations. In Proc. WWW '08, 141–
users to items through tags. In Proc. WWW '09, 671–
150. ACM (2008).
680, ACM (2009).
RecTour 2017, August 27th, 2017, Como, Italy. 39 Copyright held by the author(s).