Neural Factorization Applied to Interaction Matrix for Recommendation

Abstract

Matrix factorization methods have successfully been used for rating prediction. The interactions between users and items are recorded in the interaction matrix. In this paper, a neural matrix factorization method is proposed that is applied to the interaction matrix. More specifically, the normalized interaction matrix is given as input to the neural network in order to extract user and item embeddings. The estimated ratings are obtained by the inner product between the extracted user and item embeddings. The proposed method is assessed in movie rating prediction by employing three MovieLens datasets. It is demonstrated that the proposed neural factorization attains competitive performance and is less computationally demanding against the state-of-the-art methods in movie rating prediction.