抄録
Although recommender systems have been evaluated in accuracy to capture user satisfaction, it is argued that the bottom-line measure of the success of a recommender system should go beyond accuracy since it alone is insufficient to capture it. Techniques to enhance various aspects of recommender systems such as similarity or novelty adding to accuracy were also proposed. In this paper, we propose a recommendation method to enhance serendipity based on the assumption that a certain degree of serendipity enhances user satisfaction. The basic idea of the proposed method is that the user would be unexpected if the system recommend the user's favorite contents which are depart from the habit in access. In our method, we firstly introduce a preference model and a habit model for the user, and then estimate the serendipity based on the differences between the results by both of them. We finally make a recommendation list by merging the contents selected by the preference model and those based on the estimated serendipity. We verify the effectiveness of the proposed method through TV show recommendation.
