Nonparametric Bayesian Probabilistic Latent Factor Model for Group Recommender Systems

The explosion of the online web encourages online users to participate in group activities. Group recommender systems are essential for recommending items to a group of users based on their common preferences. However, existing group recommender systems do not exploit user interaction within a group and merely work on groups with fixed sizes of users and same levels of similarity among group members, which significantly limits its usage in real world scenarios. In this paper, we propose a novel nonparametric Bayesian probabilistic latent factor model to learn the collective users' tastes and preferences for group recommendation by exploiting user interaction within a group, which is able to well handle a variety of group sizes and similarity levels. We evaluate the developed model on three publicly available benchmark datasets. The experimental results demonstrate that our method outperforms all baseline methods for group recommendation.