Novel Recommendation based on Personal Popularity Tendency

Novel recommender systems, recommending novel items, have recently gained considerable attention in the research community. Existing novel recommender systems, however, still do not consider multiple aspects of novelty or diversify the recommendation and thus result in recommending popular items. Recommending popular items may not always satisfy users. For example, although popular movies are likely preferred by most users, they are often not very surprising or novel because users may already watched or heard about such movies. Also such recommender systems hardly satisfy movie maniacs who like to watch minor movies. This paper proposes a novel recommendation method called Personal Popularity Tendency Matching (PPTM) which recommends novel items considering individual's personal popularity tendancy (or PPT). Our analysis shows that there exist significant differences in PPTs. Our method considering PPT in recommendation helps recommendation to be diversified by reasonably penalizing popular items while minimizing performance degradation. We also propose an efficient method for matching PPTs while maximizing user preferences, based on our theoretical analyses. We experimentally show that our PPTM is better than other methods in the aspects constituting novelty.