A hybrid recommendation approach using LDA and probabilistic matrix factorization

Abstract

Recommender systems provide users with suggestions and selections. Hybrid approaches which combine the neighborhood-based methods and the model-based methods have become popular when building collaborative filtering recommenders, but similarity is established between users/items only by rating information which is just numerical value and does not contain any semantic information, leading to the loss of flexibility. To address this problem, a probabilistic matrix factorization recommendation approach fusing neighborhood selection based on Latent Dirichlet Allocation is proposed. In the proposed approach, users’ and items’ neighbors are selected through users’ interests distribution and items’ attributes distribution. Then the approach incorporates similarity matrix into probabilistic matrix factorization to obtain users’ feature vectors and items’ feature vectors to make recommendations. The experimental results show that our approach is effective to improve recommendation performance and to solve data sparsity problem.