This paper presents insights and learning experiences on the development of an integrated group recommender system inthe European FP7 HBB-Next research project. The system design incorporates insights from user re-search and evaluations, media industry players, and European HbbTV standard-ization efforts. Important differences were found between providing content recommendations for HbbTV and e.g. on-line purchases. The TV user experience is very "lean back", so the user interface and interaction has to be minimalistic. The TV broadcast schedule changes continuously, so the system has to be continuously updated. TV is typically consumed with family or friends, so it should support group recommendations. Furthermore,an important challenge is the HbbTV business ecosystem, where the content originates from multiple broadcasters and the recommendations provider may be different from the HbbTV platform provider. The resulting system is a Java-based recommender framework with open interfaces for content metadata provisioning, user-profile and identity management, group recommender algorithms, and group recom-mendation retrieval. A metadata provision system was developed, automatically enriching EPG metadata with content metadata from open Internet sources. Users are identified via QR-code scanning and face recognition. The recommender uses a genre-based collaborative “least misery” group-filtering algorithm. The client side application is an HbbTV application. Whereas most requirements could be fulfilled, further study is needed to find acceptable solutions for collecting user preferences and user identification in the HbbTV context.
展开▼
