K. Kurapati and S. Gutta (USA)
Clustering, Stereotypes, Profile Transformation, TV Recommender System, User Profiles
Personal Television is here via the advent of a new class of devices called personal video recorders (PVRs). For a PVR to provide an enriched TV experience to the user, personalization is the key. One of the thorny problems facing a recommender system is that of cold-start: how does one capture the user preferences quickly and effectively and provide user-specific personalization "out of-the-box"? To provide instant personalization, we propose a framework that allows a user to choose stereotypes, which reflect his/her TV viewing behavior closely, to seed their initial profiles and have a better "out of-the-box" experience. The stereotypes have been derived by clustering TV viewing patterns of a sample set of 7 users who have been contributing their viewing histories to us for periods ranging from 5 months to 2 years. We conducted experiments by applying randomly chosen stereotypes to users in our sample set and measured the initial recommendation errors based on ground-truth data given by the users. The best initial errors were around 30%, which compared favorably with the best errors we have achieved to date from recommenders trained on user specific data for a period of time.
Important Links:
Go Back
