Recommender System Using Community Detection and Probabilistic Matrix Factorization

The collaborative filtering approach is used in recommendation systems to identify items of interest to users based on their historical preferences. The user-item interactions are modeled into the low-rank matrices by using the probabilistic matrix factorization (PMF) technique for collaborative filtering. However, traditional PMF models do not consider the social relationships between users, which can improve the accuracy of recommendations. This paper presents a community detection-based probabilistic matrix factorization model for recommendations. We construct a user-item interaction network and discover the overlapping communities by employing the clustering technique. We compute the personalized recommendation rating score for each user by applying the PMF model to the communities in which the user is present. The performance analysis of the proposed model on MovieLens and Epinions datasets shows better performance than the simple PMF approach.