Information Technology Reference
In-Depth Information
5.5.3 User Study
In order to evaluate our recommender system we conducted a user study (with stu-
dents from our university). We analyzed the user behavior logged by the web-server
and observed the user's interactions with the system. In discussions with participants,
we recognized that a difficult problem in recommender systems is how to cope with
movies unknown to the user. On the one hand, users expect unknown, “serendipitous”
items (following the idea that a recommender should provide new suggestions); on
the other hand, users cannot rate the recommendation quality of the system, if all
recommended items are unknown. To handle this problem, the meta-recommender
(aggregating the results from the “simple” recommender agents) should provide a
diverse set of recommendations ensuring that the result set contains popular as well as
less popular entities: Relevant popular entities (probably known to the user) improve
the trust in the system. Less popular entities (probably unknown to the user) cover the
requirement of providing serendipitous movies. Moreover, the system should provide
human understandable explanations describing why a suggested movie matches the
individual user preferences. Good explanations encourage the user to accept unknown
movies as useful recommendations. Explanations generated based on semantic data
are helpful, since they describe the aspects in which a recommendation is relevant to
the user (even though the recommendation is not obvious). Additional information,
such as movie trailers or detailed movie descriptions or movie posters, is often useful
to the user giving a first impression on the suggested items. The user preferences dif-
fer fromone another. The personalized combination of different recommender agents
has been seen as an adequate approach to consider individual preference schemes
based on encyclopedic semantic recommenders.
In general, most users liked the developed approach of
facetted recommendation
giving the user many new ideas about potentially interesting movies. The visual-
ization of the recommendations encourages users exploring new facets they have
not been aware of before. Users can explore new movies relevant according to the
individual preferences. The explanations help users to understand why an unknown
movie is a relevant recommendation according to the personal profile.
5.6 Conclusion
We presented a semantic movie recommender system that overcomes the problem
of traditional recommender systems. The developed semantic recommender sys-
tem is able to aggregate different types of knowledge (rating/collaborative-based
and content-based knowledge) from heterogeneous sources. The wide variety of
integrated knowledge prevents the cold-start problem and improves the quality of
the provided suggestions. In addition, the system is extensible allowing the system
provider integrating additional knowledge resources. Since the knowledge of the
semantic recommender system is represented as one big graph (consisting of nodes
Search WWH ::
Custom Search