Movie Recommendation in Heterogeneous Information Networks

Recommender systems, as we all know, have gained tremendous popularity over the past few years and been widely used in e-commerce. Recent researches have improved recommendation performance combining additional user and item relationships with hybrid recommender systems. However, most of these studies only consider a single type of relationship while in application recommendation problems always exist in heterogeneous information networks. In this paper, we combine the model-based collaborative filtering with heterogeneous information networks to create an efficient recommendation model. We adopt meta-path to denote multiple types of entities and relationships in heterogeneous information networks and use PathSim as the similarity measurement. We employ a nonnegative matrix factorization based collaborative filtering recommendation method under each meta-path. Furthermore, we cluster users or items into subgroups and our method shows advantages through empirical studies.