Group
and Single-user Influence Modeling for
Personalized Recommendation
Michal
Kompan
Doctoral thesis
project supervised by prof. Mária Bieliková
|
Motivation and Goals
Personalized recommendations are integral part of
nowadays Web-based applications. The amount of information available to
the Web users is tremendous and increasing day by day as a result of
the information accessibility and tendency to transform Web users from
the passive role of content consumers to the active content producers.
The problem of information overload is studied a and researched from
nearly beginning of the ''information revolution''. Personalized
recommendations are generally used to overcome some of the problems
connected to information overload by reducing irrelevant or
recommending relevant information.We are facing up the tremendous
social activity over the Web increase. The popularity of social
oriented services is increasing continuously. Thanks to these trends
the need for group recommendations - recommendations suitable for the
group members is increasing.
Today's group recommenders do not or minimally considers group
structure, social connections or users personalities in order to
generate recommendations. Consideration of such extra information can
significantly improve recommendations and increase users' satisfaction
as it is clear that similar intergroup processes as in the real life
can be observed.
We target at improving recommendation approaches based
on identified problems. We propose novel methods for the personalized
recommendation for group and single users according the following aims:
- improving group recommendation performance with
respect to the user satisfaction, focusing on considering various
aspects of recommendation and users and specific domain optimization.
- improving single-user personalized recommendation by
enhancing it with group recommendation principles focusing on improving
the performance of recommendation for new users and/or specific domains
and improving the accuracy of standard recommendation approaches.
Results
In our work we focused on designing and evaluating such
approaches of personalized recommendation, which help
users in everyday life situation to overcome information
overloading problems.
Moreover, we proposed approaches for the single-user
recommendation based on the group principles by considering virtual
users or users' context influence
and the group recommendation itself.
Our results can be summarised as:
- Improvement of the group recommendation in the movie
domain by proposing voting-based recommender based for active groups.
- Improvement of the group recommendation by proposing
novel influence based recommendation method, considering users'
personalities and intergroup relationships.
-
Group recommendation improvement by proposing novel group
recommendations method for the learning task recommendation with users'
learning styles consideration.
-
Improvement of the context-aware recommendation by proposing
single-user context-aware recommender which, is based on the context to
context influence assumption.
-
Improvement of the single-user recommendation for new users by
proposing novel approach for recom mendation based on the virtual group
construction.
-
Improvement of the single-user recommendation by proposing novel
recommendation approach enhanced by virtual groups construction and
aggregation of the group preferences.
Conclusions
The consideration of social aspects influence and
personality types during the preference aggregation process (adjusting
the ratings) can bring qualitatively better recommendation for
individuals and for the group as a whole respectively. In connection to
the satisfaction modeling, various real-life scenarios can be modelled.
Not only the
horizontal influence but also the vertical influence modelling can be
benefcial in some domains (e.g., considering teachers preferences
within the group).
The thesis extended abstract is available in the Bulletin
of the ACM Slovakia.
Selected publications
- Kompan, M.,Bieliková, M.
- Social Structure and Personality Enhanced Group Recommendation. In Proc of 2nd Workshop on Emotions and Personality in Personalized Services in conjunction with UMAP 2014, Aalborg Denmark.
- Kompan, M.,Bieliková, M.
- Group
Recommendations: Survey and Perspectives. In Computing and
Informatics, ISSN 1335-9150, Vol. 33, No. 2, 2014.
- Kompan, M.,Bieliková, M.
- Context-based
Satisfaction Modelling for Personalized Recommendations. In
proceedings of the 8th International Workshop on Semantic and Social
Media Adaptation and Personalization (SMAP 2013), IEEE, pp. 33-38.
- Kompan, M.,Bieliková, M.
- Personalized
Recommendation for Individual Users Based on the Group Recommendation
Principles.
In Studies in Informatics and Control, ISSN 1220-1766, Vol. 22, No. 3,
2013, pp. 331-342.
- Bieliková, M., Kompan, M., Zeleník, D.
- Effective
Hierarchical Vector-Based News Representation for Personalized
Recommendation.
Computer Science and Information Systems Vol. 9, No. 1., ISSN
1820-0214, pp. 303-322.
- Kompan, M.,Bieliková, M.
- Content-Based
News Recommendation.. In BUCCAFURRI, F. -- SEMERARO, G. Vol.
61 E-Commerce and Web Technologies : 11 th International Conference,
EC-Web 2010 Bilbao, Spain, September 1-3, 2010 Proceedings. pp. 61-72.
|
to Homepage |
|
to Teaching |
|
to the Top |
|
|
|
|