Matrix Factorization Meets Cosine Similarity: Addressing Sparsity Problem in Collaborative Filtering Recommender System

Matrix factorization (MF) technique has been widely used in collaborative filtering recommendation systems. However, MF still suffers from data sparsity problem. Although previous studies bring in auxiliary data to solve this problem, auxiliary data is not always available. In this paper, we propose a novel method, Cosine Matrix Factorization (CosMF), to address the sparsity problem without auxiliary data. We observe that when data is sparse, the magnitude of user/item vector could not be properly learned due to lack of information. Based on that observation, we propose to use cosine to replace inner product for sparse users/items, thus eliminating the negative effects of poorly trained magnitudes. Experiments on various real life datasets demonstrate that CosMF yields significantly better results without help of auxiliary dataset.