=Paper=
{{Paper
|id=Vol-2222/paper2
|storemode=property
|title=Designing Interactive Visualizations of Personalized Review Data for a Hotel Recommender System
|pdfUrl=https://ceur-ws.org/Vol-2222/paper2.pdf
|volume=Vol-2222
|authors=Catalin-Mihai Barbu,Jürgen Ziegler
|dblpUrl=https://dblp.org/rec/conf/recsys/Barbu018
}}
==Designing Interactive Visualizations of Personalized Review Data for a Hotel Recommender System==
https://ceur-ws.org/Vol-2222/paper2.pdf
Designing Interactive Visualizations of Personalized Review
Data for a Hotel Recommender System
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 has established that online reviews can be a rich source of con-
Online reviews extracted from social media are being used increas- textual information [2, 4, 26]. When presented alongside factual
ingly in recommender systems, typically to enhance prediction product attributes and standardized ratings, reviews can provide
accuracy. A somewhat less studied avenue of research aims to additional background evidence to support users in their decision-
investigate the underlying relationships that arise between users, making process. Consequently, reviews are being used—with in-
items, and the topics mentioned in reviews. Identifying these—often creasing effectiveness—as a further means of explaining recommen-
implicit—relationships could be beneficial for at least a couple of dations [4, 20]. At the same time, large amounts of user-generated
reasons. First, they would allow recommender systems to personal- content also create an opportunity for personalization.
ize reviews based on a combination of both topic and user similarity. In this paper, we describe our ongoing approaches to personalize
Second, they can facilitate the development of novel interactive user reviews for a hotel RS and to visualize them in an interac-
visualizations that complement and help explain recommendations tive manner. The contribution of our work is threefold, namely
even further. In this paper, we report on our ongoing work to per- to: 1) propose a model for identifying a suitable set of reviews to
sonalize user reviews and visualize them in an interactive manner, show a specific user, taking advantage of implicit relationships
using hotel recommending as an example domain. We also dis- mined from those reviews; 2) develop methods to visualize said re-
cuss several possible interactive mechanisms and consider their views to support users’ decision-making; and 3) explore interactive
potential benefits towards increasing users’ satisfaction. mechanisms that allow users to maintain control over the visualiza-
tion. Our approach builds upon the co-staying concept introduced
CCS CONCEPTS in [1], wherein implicit multimode (user-topic-item) relationships
extracted from user-generated content may be useful for increasing
• Information systems → Recommender systems; Personal-
the trustworthiness of hotel recommendations.
ization; • Human-centered computing → User centered design;
In the following section, we report on the state of the art in review
personalization and in information visualization techniques for RS.
KEYWORDS Afterwards, we present our conceptual model for personalizing
Recommender systems, Personalized reviews, Interactive visualiza- reviews, using hotel recommendations as an example domain. Sub-
tion, Tourism, Multimode networks, Trustworthiness sequently, we propose an approach for visualizing the data based
on Sankey diagrams [24]. We also describe several mechanisms
1 INTRODUCTION AND MOTIVATION for interacting with the visualizations. Finally, we conclude by re-
flecting on our approach and enumerating promising directions for
As the research focus in recommender systems (RS) shifts gradually
future research.
from prediction accuracy towards more user-centric methods, top-
ics such as personalizing the user experience and increasing users’
trust become more salient [11]. Transparency [18] and control [6] 2 RELATED WORK
are frequently mentioned in the literature as important factors for Although the importance of online reviews for explaining recom-
achieving these goals. In this context, various approaches have mendations has been recognized in prior work (see, e.g., [20] for
been developed to support users in their exploration of recommen- an overview), the topic of personalizing the presentation of re-
dations. Collectively, these approaches are usually referred to as views in RS has received relatively little attention from researchers.
interactive recommending [7]. Moghaddam et al. [14] provides empirical evidence to support the
When many attributes need to be considered before making a fact that the perceived quality and helpfulness of online reviews
choice, as is often the case in hotel RS, comparing ranked lists of rec- differs across users. Their evaluation, which was performed on a
ommendations often becomes cumbersome [3]. At the same time, real-life dataset of reviews, compared two latent factor models for
alternative visualization techniques need to strike a fine balance predicting the personalized review quality. Similarly, Tu et al. [21]
with respect to the amount of information that can be presented aim to personalize the set of reviews shown to users by decreasing
while maintaining ease of understanding. Because of this inher- redundancy and maximizing the coverage of topics of interest. Once
ent difficulty, ranked lists are frequently, despite their shortcom- a suitable set of reviews has been identified, the next challenge is
ings, the preferred way to display recommendations. A promising how to present them.
middle-ground approach is to visualize specific aspects of a rec- Information visualization for RS is an active and promising field
ommendation (e.g., user-generated content) while still retaining of research [10]. Several approaches have been proposed for vi-
the traditional presentation style for the item lists. Prior research sualizing recommendations in an interactive manner. We believe
RecTour 2018, October 7th, 2018, Vancouver, Canada. 7 Copyright held by the author(s).
some of these approaches could also be adapted for visualizing
specific aspects of a recommendation. In SetFusion [16], a hybrid RS
for conference talks, the authors enhance the typical ranked list
paradigm with interactive Venn diagrams. The charts afford users
a new perspective on examining and filtering recommendations.
The implementation is a successor of TalkExplorer [23], in which
the relevant information was represented using cluster maps. Yazdi
et al. [25] propose a bubble graph representation for suggesting
collaboration opportunities. They show that the visualization helps
users form a mental model of the recommendation space and the
connections between scholars, institutions, and research topics. A
similar visualization metaphor is also used in [15] to recommend
contacts in social networks. Richthammer and Pernul [17] employ
treemapping to facilitate users’ exploration of movie recommenda-
tions. They show that the structured presentation makes it easier
for users to obtain an overview of the search space and possible Figure 1: Eliciting user preferences. Users can drag and drop
alternatives. In contrast, Kunkel et al. [12] render the movie domain relevant topics from the categories on the left-hand side to
space on a 3D map that can be reshaped by users to “uncover” sim- the “Preferences” area on the right-hand side. Sliders can be
ilar recommendations. Finally, Tietz et al. [19] proposes a method used to adjust the importance of each attribute.
for visualizing multimedia content based on linked data. Displaying For the sake of simplicity, and as an initial step towards testing
the semantic relationships graphically supports exploration and our hypothesis, we decided to elicit user preferences as part of the
the discovery of new content. recommendation process. Concretely, in our application—which is
Despite the multitude of techniques, most of them are inherently based on the one described in [5]—users shall be asked to select (and
limited in the number of elements that can be realistically depicted assign weights to) hotel characteristics that are most important to
on a screen. Thus, identifying and grouping items into clusters them (Figure 1). This interaction bears similarities to how a person
becomes a key requirement for reducing clutter and helping users typically interacts with online booking portals: After choosing a
cope with the amount of information. Several approaches have been destination and travel date, users are normally presented with a list
proposed in the field of social network analysis that can be applied of filters that they may use to refine the list of recommendations
to multimode networks (see, for instance, [8], [9], and [13]). even further. Clicking on a filter labeled “beach”, for instance, will
prioritize hotels located near the seafront. Such an action can be
3 PERSONALIZE A SET OF HOTEL REVIEWS regarded as preference elicitation. In our prototype, we feed this
Whether a hotel review is considered helpful by a user may de- information into the RS not only to find recommendations, but also
pend on several aspects, among them individual preferences (e.g., to personalize the reviews.
“I prefer to sleep on a soft mattress; what have previous guests Once they have been elicited, user preferences can be matched
written concerning bed quality?”), the specifics or requirements of against pre-extracted topics (see [5]) to select the most suitable
the travel scenario (e.g., “I am traveling for work, so I am mostly reviews. For each review belonging to one of the recommended
interested in the opinions of other business travelers.”), and various hotels, a partial relevance score, R c , can be computed based on
sociodemographic factors (e.g., “What do people who, like me, usu- the number of topics that match the user preferences. A second,
ally book 3-star hotels think about these accommodations?”). The and arguably more interesting step, is to additionally consider user
goal of personalization is to show users the most relevant reviews, similarity when calculating a review’s final relevance score. We
based on their recorded preferences [21]. Our hypothesis is that identified four user factors that we consider relevant for this task.
both the content of the review and metadata about the person who A reviewer’s rating behavior denotes the extent to which her hotel
wrote it can be leveraged to calculate a relevance score. This would scores match those of other users who share similar preferences.
allow a RS to prioritize hotel reviews that: 1) mention the topics in This is, in essence, the basis for collaborative filtering [11]: For a
which the user is interested; and, at the same time, 2) are written given set of hotels, we expect like-minded guests to give more ho-
by people who have the most in common with the user. mogeneous ratings. The travel profile represents a combination of
Various techniques have been proposed for extracting features aspects that characterize the reviewer’s typical hotel booking. These
and user attitudes from online reviews [2, 4, 26]. Most commonly, may include the purpose of travel (i.e. business or leisure), room
the output is a list of concepts, or topics, that appear often in re- type, number of nights, time of year etc. Another factor is the de-
views (for example, “soft bed” or “quiet room”). User attitudes about gree to which a reviewer’s own set of preferences is well-defined. For
a certain topic can be either positive, negative, or neutral [26]. In [1], example, reviews contributed by someone who often gives feedback
we described how the connections between users, hotels, and topics on the quality of the bed are probably more relevant to a user who
form an implicit social network—meaning that users do not com- cares about this aspect of a hotel room. Finally, we check whether
municate directly with each other. Instead, relationships are formed the reviewer has stayed in similar hotels. For this, we consider both
based on the hotels that they have visited in the past and the topics objective information, such as a hotel’s star rating, and prevalent
that they mentioned in their reviews. topics extracted from user-generated content. Prior work suggests
RecTour 2018, October 7th, 2018, Vancouver, Canada. 8 Copyright held by the author(s).
Figure 2: Proposed model for calculating the relevance score of a review, taking into account both its content and its author.
that people who book similar hotels may also have comparable a more visual manner. Concretely, we started developing graphical
expectations [2, 21]. By combining these factors, the second partial representations of relevant hotel topics (and their authors) based
relevance score, R u , can be calculated. The review’s final relevance on: 1) how often they appear in the user-generated content; and
score can be written as R r = R c · wc + R u · (1 − wc), where the 2) their valence (i.e. positive or negative mentions). To avoid infor-
weighting factor wc will be found empirically. An overview of the mation overload, we purposefully restrict the visualization to only
proposed model is shown in Figure 2. With the exception of travel a personalized set of hotel reviews, as identified in the previous
profile, all factors can be extracted from information contained section. Our aim is to find out whether such a visualization has a
in the co-staying network. The remaining factor can be obtained significant effect in terms of helping users understand better why a
from the reviews’ metadata (i.e. the review date and automatically- hotel was recommended. Thus, we consider the visualization as an
generated tags about the hotel booking, such as duration of stay, additional form of explanation. Constraining the visual represen-
type of room, and number of guests). The dataset used for generat- tation to relatively small amounts of data (i.e. from a personalized
ing the co-staying network is the one described in [1]. subset of reviews) also alleviates the main shortcoming identified in
As a further refinement, we will also explore the possibility of the related work section. At the same time, we believe our approach
using reviews written for hotels that are part of the same chain as remains in line with the typical use cases of hotel RS. Specifically,
the recommended hotels. Our premise is that hotel chains typically most people have a limited number of preferences (i.e. topics) in
strive to achieve a consistent user experience across their sites [1]. which they are interested in for a given trip.
This means that, per our co-staying concept, two reviewers can We experimented with two graphing methods, namely: 1) Treemap,
be considered similar even if they previously booked rooms in an area-based visualization [17]; and 2) Sankey, a type of flow dia-
different locations of the same hotel franchise. We aim to evaluate gram [24]. Both techniques have specific advantages and shortcom-
our approach by comparing it against latent factor models, such as ings. In general, Treemaps provide a good overview, but users might
the one suggested in [14]. We believe the additional relationships find it more difficult to focus on specific details. In contrast, Sankey
captured by the multimode network will yield improved results diagrams tend to have a higher legibility. This is due to their flow
when compared to other review personalization approaches. structure, which generally follows a left-to-right (or, less frequently,
top-to-bottom) orientation that might be easier for users to grasp.
4 VISUALIZE AGGREGATED REVIEW DATA Because of this aspect, we will focus on Sankey visualizations in
Based on our review of the literature (see section 2), we believe there the remainder of this paper.
is significant potential in combining traditional RS with a means to The layout of a Sankey diagram is flexible enough to accom-
explore information related to a specific hotel recommendation in modate multiple levels of nodes. As a result, it is well-suited for
RecTour 2018, October 7th, 2018, Vancouver, Canada. 9 Copyright held by the author(s).
(a) (b)
(c) (d)
Figure 3: Example visualizations using Sankey diagrams. Different colors (green and red) and symbols (“+” and “-”) are used to
denote positive and negative mentions, respectively. Top left: Topics mentioned by two users in their reviews about Hotel A.
Top right: Topics mentioned by a user in her hotel reviews. Bottom left: Opinions regarding the location of two hotels have
been aggregated based on users’ travel category. Links originating from the group “business travelers” are highlighted. Bottom
right: Subset of topics mentioned by a group of users who reviewed Hotel A.
visualizing multidimensional data, such as the user-topic-hotel re- the amount of information that is represented in the chart. One way
lationships that form the backbone of our co-staying network. Four to achieve this is by clustering nodes to reduce clutter and increase
typical visualizations are shown in Figure 3. Each follows a similar legibility. This is especially relevant in the case of user nodes, which
pattern, with the user (or user group) nodes placed on the left, topic will almost always be the most numerous of the three vertex types.
nodes in the middle, and hotel nodes on the right. Edges between A relatively straightforward possibility is to group users based
nodes correspond to topic mentions; the width of an edge is pro- on whether they are traveling for business or leisure (Figure 3c).
portional to the number of times its corresponding topic appears A more interesting approach that we are investigating is how to
in a user’s reviews. User sentiment is represented using colors (i.e. cluster users based on their similarity scores, which are computed
red and green for negative and positive mentions, respectively) using the algorithm described in the previous section. Furthermore,
and symbols (i.e. “-” for negative and “+” for positive mentions). topics can also be clustered, for example based on whether they
Furthermore, the coloring of topic and hotel nodes indicates the refer to the hotel in general (e.g., “location”), a room feature (e.g.,
proportion of positive vs. negative references. These graphical ele- “shower”), or the quality of the service (e.g., “staff”).
ments are meant to help users perceive quickly the prevailing user Users will also have the option to “zoom” in or out in order to fine
sentiment on a given issue. Specific paths in the Sankey diagram tune the level of detail. Another way to control the visualization is
can be highlighted to increase their salience, as shown in Figure 3c. by providing adequate filtering mechanisms. For example, the user
As depicted in Figure 3c and Figure 3d, the visualization can also may select only a subset of topics to visualize, or she might decide to
be used to compare two or more hotels. view only topics with negative opinions. Even so, showing all three
Since many prospective users might not be familiar with Sankey layers of the underlying multimode network at once might still
diagrams, we formulate several interactive mechanisms to support prove too difficult for some users to comprehend. Therefore, one
them. First, and most importantly, users should be able to control possible solution is to limit the visualization to only two types of
RecTour 2018, October 7th, 2018, Vancouver, Canada. 10 Copyright held by the author(s).
(a) (b)
Figure 4: Reviews are displayed (on demand) as a separate layer on top of the Sankey visualization. Topics are highlighted
according to their valence. Left: Users’ opinions about a particular topic related to Hotel A. For the top review in the list, only
a relevant snippet is shown. Right: Partial view of the reviews written by a user.
vertices. In this case, suitable interface elements could be provided 5 DISCUSSION AND FUTURE WORK
to facilitate interaction with the third dimension, e.g., by using As the amount of user-generated content continues to grow, it is be-
filters. coming increasingly important to develop methods for filtering and
Clicking on the nodes also affords interesting interaction oppor- personalizing the content used for explaining recommendations.
tunities. One example is to allow users to “refocus” the visualization We propose a model for identifying personalized sets of reviews
around a specific node. In Figure 3a, clicking on one of the two in a hotel RS, which combines both content and user similarity
users changes the diagram to show only the topics mentioned by to calculate a relevance score for each review. In particular, we
that user (Figure 3b). Similarly, selecting a topic would display only believe that better user similarity measures can be developed by
the users who referred to that topic in their reviews. Finally, click- taking into account ternary relationships such as those in our co-
ing on the hotel would have the effect of reverting to the default staying network [1]. Specifically, we are investigating connections
visualization. An interesting open question, which we plan to verify between travelers who: 1) booked the same hotel(s); 2) stayed in
empirically, is whether to allow users to reorganize the diagram similar hotels (e.g., that are part of the same chain); 3) have a well-
by dragging and dropping nodes. Such functionality may facilitate defined set of topics that they mention frequently in their reviews;
“ad-hoc” clustering. Moreover, the resulting arrangement could also and 4) exhibit a similar rating behavior. Furthermore, we suggest a
be saved as a template, so that future visualizations are rendered, method for displaying a subset of personalized reviews graphically
by default, in a similar fashion. using Sankey diagrams. Allowing users to explore the multimode
Initially, our Sankey diagram implementation does not display relationships could be considered as an additional form of explain-
the actual content of the reviews. However, users can easily access ing recommendations [20]. As future work, we aim to evaluate
this information on demand (cf. Figure 4). One relatively simple empirically whether these approaches, combined, increase users’
method to achieve this functionality is to render the appropriate understanding of the reasons behind recommending a specific hotel.
reviews in an overlay window. The content and presentation style We expect that such an outcome would, in turn, have a positive ef-
are determined by the node or edge with which the user interacted. fect on the transparency and perceived trustworthiness of hotel RS.
In Figure 4a, interacting with the node “staff”—e.g., by double- Although not specifically discussed in this paper, methods for
clicking—displays users’ feedback on that topic. (Note that the visualizing user opinions could be of interest also to hotel managers.
underlying Sankey diagram is identical to the one in Figure 3a.) In combination with interactive mechanisms, such as the ones
Furthermore, the top review in the aforementioned example has suggested in the previous section, these graphical representations
been condensed to a relevant snippet; however, the user may tog- could provide a clearer picture of the feedback that guests typically
gle an embedded link to view the entire text. By the same token, write. This could help monitor and focus on areas that require
interacting with either a hotel or with a user node depicts all hotel improvement, i.e. topics with numerous negative mentions. The
reviews, or the opinions contributed by a specific user, respectively. usefulness of these methods in other domains, such as data analytics
An example of the latter is shown in Figure 4b (see also Figure 3b or visualization RS [22], should also be investigated further.
for the initial visualization). Moreover, this type of interaction is
implemented for edges as well. Alternatively, a user may only be
interested in finding out quickly how many times a topic has been ACKNOWLEDGMENTS
mentioned, without perusing the reviews. In this case, simply hov- This work is supported by the German Research Foundation (DFG)
ering over an edge will display this information in a summarized under grant No. GRK 2167, Research Training Group “User-Centred
form, e.g., “‘breakfast’ → 5 mentions (mostly positive)”. Social Media”.
RecTour 2018, October 7th, 2018, Vancouver, Canada. 11 Copyright held by the author(s).
REFERENCES [14] Samaneh Moghaddam, Mohsen Jamali, and Martin Ester. 2011. Review recommen-
[1] Catalin-Mihai Barbu and Jürgen Ziegler. 2017. Co-Staying: A social network dation: personalized prediction of the quality of online reviews. In Proceedings of
for increasing the trustworthiness of hotel recommendations. In RecTour ’17. the 20th ACM International Conference on Information and Knowledge Management.
CEUR-WS Vol. 1906, 35–39. ACM, 2249–2252.
[2] Guanliang Chen and Li Chen. 2015. Augmenting service recommender systems [15] Sayooran Nagulendra and Julita Vassileva. 2014. Understanding and controlling
by incorporating contextual opinions from user reviews. User Modeling and the filter bubble through interactive visualization: a user study. In Proceedings
User-Adapted Interaction 25, 3 (2015), 295–329. of the 25th ACM International Conference on Hypertext and Social Media. ACM,
[3] Cecilia di Sciascio, Vedran Sabol, and Eduardo E Veas. 2015. uRank: Exploring 107–115.
document recommendations through an interactive user-driven approach.. In [16] Denis Parra, Peter Brusilovsky, and Christoph Trattner. 2014. See what you want
IntRS ’15. CEUR-WS Vol. 1438, 29–36. to see: visual user-driven approach for hybrid recommendation. In Proceedings of
[4] Tim Donkers, Benedikt Loepp, and Jürgen Ziegler. 2018. Explaining recommen- the 19th International Conference on Intelligent User Interfaces. ACM, 235–240.
dations by means of user reviews. In ExSS ’18. CEUR-WS Vol. 2068. [17] Christian Richthammer and Günther Pernul. 2016. Explorative analysis of rec-
[5] Jan Feuerbach, Benedikt Loepp, Catalin-Mihai Barbu, and Jürgen Ziegler. 2017. ommendations through interactive visualization. In International Conference on
Enhancing an Interactive Recommendation System with Review-based Informa- Electronic Commerce and Web Technologies. Springer, 46–57.
tion Filtering. In IntRS ’17. CEUR-WS Vol. 1884, 2–9. [18] Rashmi Sinha and Kirsten Swearingen. 2002. The role of transparency in recom-
[6] F Maxwell Harper, Funing Xu, Harmanpreet Kaur, Kyle Condiff, Shuo Chang, mender systems. In CHI’02 Extended Abstracts on Human Factors in Computing
and Loren Terveen. 2015. Putting users in control of their recommendations. In Systems. ACM, 830–831.
Proceedings of the 9th ACM Conference on Recommender Systems. ACM, 3–10. [19] Tabea Tietz, Jörg Waitelonis, Joscha Jäger, and Harald Sack. 2014. Smart Media
[7] Chen He, Denis Parra, and Katrien Verbert. 2016. Interactive recommender Navigator: Visualizing recommendations based on linked data. In International
systems: A survey of the state of the art and future research challenges and Semantic Web Conference (Industry Track). CEUR-WS Vol. 1383.
opportunities. Expert Systems with Applications 56 (2016), 9–27. [20] Nava Tintarev and Judith Masthoff. 2015. Explaining recommendations: Design
[8] Dmitry I Ignatov, Alexander Semenov, Daria Komissarova, and Dmitry V and evaluation. In Recommender Systems Handbook. Springer, 353–382.
Gnatyshak. 2017. Multimodal Clustering for Community Detection. In For- [21] Wenting Tu, David W Cheung, and Nikos Mamoulis. 2017. More focus on what
mal Concept Analysis of Social Networks. Springer, 59–96. you care about: Personalized top reviews set. Neurocomputing 254 (2017), 3–12.
[9] Isaac Jones, Lei Tang, and Huan Liu. 2015. Community discovery in multi-mode [22] Manasi Vartak, Silu Huang, Tarique Siddiqui, Samuel Madden, and Aditya
networks. In User Community Discovery. Springer, 55–74. Parameswaran. 2017. Towards visualization recommendation systems. ACM
[10] Bart P Knijnenburg, Martijn C Willemsen, Zeno Gantner, Hakan Soncu, and SIGMOD Record 45, 4 (2017), 34–39.
Chris Newell. 2012. Explaining the user experience of recommender systems. [23] Katrien Verbert, Denis Parra, and Peter Brusilovsky. 2014. The effect of different
User Modeling and User-Adapted Interaction 22, 4-5 (2012), 441–504. set-based visualizations on user exploration of recommendations. In IntRS ’14.
[11] Joseph A Konstan and John Riedl. 2012. Recommender systems: from algorithms CEUR-WS Vol. 1253, 37–44.
to user experience. User Modeling and User-Adapted Interaction 22, 1-2 (2012), [24] Krist Wongsuphasawat and David Gotz. 2012. Exploring flow, factors, and out-
101–123. comes of temporal event sequences with the outflow visualization. IEEE Transac-
[12] Johannes Kunkel, Benedikt Loepp, and Jürgen Ziegler. 2017. A 3D item space tions on Visualization and Computer Graphics 18, 12 (2012), 2659–2668.
visualization for presenting and manipulating user preferences in collaborative [25] Mohammad Amin Yazdi, André Calero Valdez, Leonhard Lichtschlag, Martina
filtering. In Proceedings of the 22nd International Conference on Intelligent User Ziefle, and Jan Borchers. 2016. Visualizing opportunities of collaboration in large
Interfaces. ACM, 3–15. research organizations. In International Conference on HCI in Business, Government
[13] Bo Long, Zhongfei Mark Zhang, Xiaoyun Wu, and Philip S Yu. 2006. Spectral and Organizations. Springer, 350–361.
[26] Yongfeng Zhang. 2015. Incorporating phrase-level sentiment analysis on tex-
clustering for multi-type relational data. In Proceedings of the 23rd International
tual reviews for personalized recommendation. In Proceedings of the 8th ACM
Conference on Machine Learning. ACM, 585–592.
International Conference on Web Search and Data Mining. ACM, 435–440.
RecTour 2018, October 7th, 2018, Vancouver, Canada. 12 Copyright held by the author(s).