BDMF: A Biased Deep Matrix Factorization Model for Recommendation

Abstract

As a representative collaborative filtering method, matrix factorization has been widely used in personalized recommendation. Recently, deep matrix factorization model, which utilizes deep neural networks to project users and items into a latent structured space, has received increased attention. In this paper, inspired by the idea of BiasedSVD that introduces bias to both users and items, we propose a novel matrix factorization model with neural network architecture, named BDMF, short for Biased Deep Matrix Factorization. Specifically, we first construct a user-item interaction matrix with explicit ratings and implicit feedback, and randomly sample users and items as the input. Next, we feed this input to the proposed BDMF model to learn latent factors of both users and items, and then use them to predict the ratings for personalized ranking. We also formally show that BDMF works on the same principle as BiasedSVD, which means that BDMF can be viewed as a deep neural network implementation of BiasedSVD. Finally, extensive experiments on real-world datasets are conducted and the results verify the superiority of our model over other state-of-the-art.