An Effective Approach to TV Show Recommendation

Abstract

TV show recommendation is to recommend preferable TV shows in real-time at the moment the user is watching TV. Since the number of channels are increasing, TV show recommender system has become important. TV show domain has distinct characteristics to take in account for building an effective TV show recommender system. The characteristics are based on the time factor as well as the configuration of the TV show. The time factor indicates that the TV show and the user are affected by time. In terms of the TV show, it has different broadcasting time. Thus, depending on the time, the list of TV shows that are being broadcasted may vary. Also, TV show user has different watching time. This indicates that the available time the user can watch a TV show may be different according to the time the user is watching TV. Moreover, the time factor implies that the user has to select one TV show from among all shows that are being broadcasted. It is because the user cannot watch a TV show after the show ends the broadcast. Therefore, the user must select the most preferable TV show from the TV shows that are being broadcasted. For the configuration of the TV show, it indicates that TV shows consist of several episodes. It enables the user to watch a TV show repeatedly. Therefore, building a TV show recommender system imposes several challenges; (1) estimate various implicit feedback the user expresses on the TV show (2) predict the user preference of the TV show in which the user has not expressed a feedback. In this paper, we propose a method that captures user’s various implicit feedback with the consideration of the time factor and the configuration of the TV show. Also the method accurately predicts the user preference of the entire TV shows through an effective matrix factorization technique. In result, extensive experiment has been conducted and the result shows that our method outperforms all existing methods in recommendation accuracy.