Time-Dependent User Profiling for TV Recommendation

TV is one of the most important sources of media content consumption. The large amount of TV channels and programs have overwhelm audiences. It poses difficulties for viewers in finding their preferred programs. Tools for searching TV programs such as TV guides and PreVue channel are designed for general public and do not provide personalized recommendation. Developing an effective recommender system for TV is challenging because a TV is often shared by multiple people (e.g., family members) without login, and thus it is hard to acquire individual TV watch log, which is crucial to build an effective recommendation. Existing recommender systems for social networks or web commerce are devised for handling one user per account, and thus are not proper for TV recommender system. This paper proposes a time dependent user profiling technique. Particularly, we do time based analysis in which we first split watch log into certain time slots, and re-merge consecutive time slots by using a clustering technique. Evaluation results show that the proposed method produces higher accuracy than a typical profiling technique.