An adaptive preference learning method for future personalized TV

This paper presents an adaptive preference learning method based on TV-Anytime metadata for future personalized TV. To reduce CPU cost, a KWL (keyword, weight, like-degree) structure is proposed for user preference modeling. The user profile is updated dynamically based on both explicit and implicit feedback information from the user. The system provides recommendations according to the similarity rank of the programs. An experiment shows that the proposed method can learn the user's preferences efficiently.