Kaushal Kurapati

Exposing profiles to build trust in a recommender

This paper describes a method for increasing trust in a TV show recommender. We look for people in common between programs users watch and new programs that are highly rated by our TV show recommender. We then present these to users in a conversational sentence, helping them decide if they want to try the new show. This method has been implemeted in our current TV show recommender interface and will be tested in the near future.

From Stereotypes to Personal Profiles Via Viewer Feedback

For a Personal Video Recorder (PVR) to provide an enriched TV experience to the user, personalization, achieved via a recommender engine, is the key. One of the thorny problems facing a recommender system is that of a cold start: how does one capture the user preferences quickly and effectively and provide user-specific personalization? To address the cold start problem, we propose a stereotype-enabled personalization framework that transforms a user’s TV profile from a stereotypical, initial profile, to a personalized, more relevant one. The stereotypes have been derived from a sample set of users who have been contributing their TV viewing histories for periods ranging from five months to two years. We conducted three sets of experiments, each with a different set of stereotypes applied to all the users. The best performance is in the range of 11% error. This performance compares favorably with the best we have got to date on recommenders trained on user-specific data.