=Paper=
{{Paper
|id=None
|storemode=property
|title=Voting Based
Group Recommendation: How Users Vote
|pdfUrl=https://ceur-ws.org/Vol-1210/SP2014_08.pdf
|volume=Vol-1210
|dblpUrl=https://dblp.org/rec/conf/ht/KompanB14
}}
==Voting Based
Group Recommendation: How Users Vote==
https://ceur-ws.org/Vol-1210/SP2014_08.pdf
Voting Based Group Recommendation: How Users Vote
Michal Kompan Mária Bieliková
Slovak University of Technology Slovak University of Technology
Inst. of Informatics and Software Engineering Inst. of Informatics and Software Engineering
Ilkovičova 2, 842 16 Bratislava, Slovakia Ilkovičova 2, 842 16 Bratislava, Slovakia
name.surname@stuba.sk name.surname@stuba.sk
ABSTRACT 2. VOTING BASED RECOMMENDATION
It has been shown that social information as group structure Proposed approach consist of the construction of user’s rat-
or personality characteristics improve the group recommen- ings matrix, which is created based on users’ votes (Items x
dation. Sometimes no such information is available, specifi- Votes). Every user can vote for the items already voted by
cally when ad-hoc groups are constructed. Moreover, often other users, or the new item can be added as the suggestion
the items’ content is not available (or users’ preferences are to the group. Next, the matrix of normalized ratings is con-
unknown). In this paper we explore the usage of voting structed (Min-max normalization) in order to minimize low
based group recommendation and the users preference for or hight ratings influence to aggregation strategy. Finally,
such a method settings – we analyze aggregation strategies the total of three representative aggregation strategies (addi-
preferences, sharing preferences and users re-rating consis- tive, multiplicative and additive with minimal satisfaction)
tency. are used in order to construct the group recommendation,
which is presented to users:
Categories and Subject Descriptors
H.3.3 [Information Technology and Systems]: Informa- 1. Create user’s rating matrix and the normalized rating
tion filtering matrix respectively.
General Terms 2. Aggregate votes from group members (users rating ma-
Experimentation, Human Factors trix).
3. Recommend items with highest votes.
Keywords
Group recommendations, voting, aggregation strategies
Not only the lack of users’ preferences knowledge or suffi-
cient group activity indicate to use the voting based group
1. INTRODUCTION recommendation. Often there is no information about the
Group recommendation gets more and more attention in
recommended content available (e.g., movie genre, director),
today’s adaptive web-based applications [1]. Users’ social
which are used for the standard similarity search. In the
activity over the Web is increasing and thus new domains
voting based approach, this information is processed by the
and applications as movie, learning or games are available.
users, thus no content analysis or the lack of new items is
When recommending to the group of users the social struc-
required or present.
ture and personal characteristics plays important role from
the group satisfaction point of view [3]. On the contrary,
sometimes there is not possible to obtain these characteris- 2.1 Evaluation and Results
tics. When the group is constructed ad hoc – from “random” Proposed approach was implemented as a simple web-based
users it is almost impossible to collect information about the application MovieRec and available for the free usage within
group structure or users characteristics (usually obtained by the social network Facebook during the experiment. We ex-
various questionnaires) [2]. pected that – users’ ratings are more consistent as when
no sharing preferences are presented. We also believe that
One of the best performing approaches for the group rec- users’ ratings are influenced by the group context – users’
ommendation, which is suitable for active groups is the rec- re-ratings (rating previously rated item in new event and
ommendation based on voting of group members. Group group) are influenced by the group and event context. The
members suggest their preferred items and then the voting total of 73 real users within 10 days voted for 902 movies (ob-
is performed by the group. It is clear that the voting pro- tained from IMDB database), which were self-divided into
cess, especially when performed online and when the goal is the 11 groups and 93 voting events.
to reach consensus, can be influenced and enhanced by vari-
ous aspects (e.g., sharing preferences, aggregation strategies, The task presented to the users was to create or to join
group size, users’ consistency). In order to investigate the some event and try to reach consensus (based on the vot-
influence of these specific aspects we propose a voting mech- ing) on which items should be watched together within the
anism in the domain of movies. group. For every created event the users voted for their
Table 1: Voting strategies comparison.
Strategy Winning events SD Avg. vote
Additive 184 0.90 4.14
Multiplicative 147 0.83 4.08
Additive(LM) 138 0.95 4.12
a large group is interacting. Obtained results clearly show
that when a large group is requesting for the recommenda-
tion, the minimal satisfaction from the group point of view
Figure 1: Ratio of winning voting strategies com- decreases the quality of recommendation. This is supported
pared to the group size. by the standard deviation of obtained votes for particular
strategies (Table 1). From the average score point of view,
the additive strategy with least misery outperforms the mul-
tiplicative, thus the preference diversity was probably small
candidates to watch. They could create new suggestions un- within the group members.
til the event deadline. During the experiment we were ob-
serving the users’ behavior based on the sharing preferences Results – users’ consistency. Finally, we investigated
(in the half of events the preferences of other group mem- users’ consistency over the various voting and events. We
bers were visible), users’ consistency and the performance of focused on movies rated by the user in some event and
used aggregation strategy. After the event deadline, three his/her rating for the same movie in other events. In or-
lists of the generated recommendations were presented to der to minimize users’ effort, if the movie was rated by the
every user of the group (additive, multiplicative and the ad- user before, we presented this rating as default value (and
ditive with minimal satisfaction consideration strategy). Ev- the user was able to adjust this rating). The total of 462
ery user rated for the best recommendation of these three such “re-ratings” were given by the users, while only in 71
presented lists. occurrences the users changed the value of previous rating.
This is an interesting result, which can be partially caused
Results – aggregation strategies. Our first question was by the pre-filled ratings. On the other hand, the proportion
which strategy is preferred based on the group size. When of users which were consistent (85%) indicates that users
comparing the winning strategy depending on the group size adjust their ratings to the actual group context minimally
we discovered that larger groups (more single-users’ prefer- (which is supported by the social psychologist as the ten-
ences have to be aggregated) prefer additive strategy, while dency to act consistent in various situations.
the decreasing trend can be observed when multiplicative
strategy is used (Figure 1). Finally, the additive strategy 3. CONCLUSIONS
with least misery performed the worst. This can be ex- When there is no additional information about the group
plained by the fact that least misery prefers votes from the available, the voting strategy seems to be the optimal solu-
minority, thus when only one user dislikes an item, this tion. Here, the recommendation task is moved to the group
item will not be recommended. With the group size and members directly. As we shown the additive and multiplica-
users’ satisfaction, the number of such users is increasing, tive strategy are more preferred by small groups, while on
thus the quality of recommendation is decreasing. Similarly, the other side for larger groups the additive strategy is pre-
when the multiplicative strategy is used, low ratings of few ferred. Proposed voting approach revealed that the sharing
users can influence whole recommendation dramatically, ob- preferences have no or minimal influence to the group mem-
tained results supports this hypothesis – the additive strat- bers in adjusting their preferences.
egy within large groups balances the influence of deviating
individuals and the rest of members.
4. ACKNOWLEDGMENTS
Results – sharing preferences. Next, we focused on in- The authors wish to thank Ján Trebul’a for helping with
fluence of sharing preferences. Users’ events were divided implementation of MovieRec. This work was partially sup-
into the two sets – users who saw preferences of their col- ported by the grants No. VG1/0675/11 and APVV-0208-10.
leagues, and second set, where no sharing preferences were
displayed. We discovered that the sharing preferences do not 5. REFERENCES
have (or have very small) influence on the user’s ratings. The [1] M. Kompan and M. Bielikova. Group
standard deviation of these two groups differs only 0.0212. Recommendation: Survey and Perspectives. In
Thus, we see that the users in our experiments considered Computing and Informatics, 33(2):446–476.
the preference of others minimally, or were very consistent [2] M. Gartrell, X. Xing, Q. Lv, A. Beach, R. Han,
in their similar opinions and thus sharing preferences were S. Mishra, and K. Seada. Enhancing group
redundant. recommendation by incorporating social relationship
interactions. In Proc. of the 16th ACM Int. Conf. on
In general, the winner, in the most of events is the addi- Supporting Group Work, pages 97–106. ACM, 2010.
tive strategy, followed by the multiplicative and the additive [3] J. Masthoff and A. Gatt. In pursuit of satisfaction and
with minimal satisfaction strategy. This is quite surprising the prevention of embarrassment: Affective state in
result, while the minimal satisfaction seems to be not so de- group recommender systems. UMUAI, 16(3-4):281–319,
sirable (from the majority points of view), especially when Sept. 2006.