Semantic Movie Recommendations - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 139

actor director writer composer costume

designer

RDF RDF RDF RDF RDF

Fig. 5.4 The movie recommender system is implemented as an agent-based system. Each semantic
relationship set is wrapped by one recommender agent tailored to the specific properties of the
respective semantic dataset. The suggestions from the recommender agents are aggregated taking
into account the individual user preferences

recommendations, there is one agent for each user managing the individual user
profile and the selection of the recommender agents for each request.
Incoming user requests are handled by the web server. When a user requests
recommendations, the web server delegates the task to the personal agent. This agent
sends the request to all recommender agents relevant for the current user. The results
from the recommender agents are collected and aggregated into a final result list.
Finally, the aggregated list (might be filtered by user-defined criteria, such as motion
picture rating or popularity ) annotated with additional information is presented to
the user. For each suggested movie the system provides an explanation describing
the semantic relations between the suggested movie and the user's favorite movies
(explicitly defined by the user).

5.5.2 The Graphical User Interface

Due to the fact that the developed recommender system is based on a semantic graph,
users must define the preferences by selecting preferred entities (graph nodes). Our
system handles the problem by suggesting users the entities matching best the user
input. This approach efficiently supports the users in finding the preferred entities and
avoids problems with ambiguous entities. Figure 5.5 shows an example for the auto-
completion , suggestingmoviesmatching the user input based on the textual similarity.
Based on the defined query the system calculates the recommendations con-
sidering several different semantic relationship sets. Thus, it computes the entities
most strongly related according to the semantic relationship sets. Our movie recom-
mender system aggregates nine semantic relationship sets: Actors , Directors ,
Misc (e.g., stuntman, location scouts, and caterer), Composers , Producers ,