Using Approximately Coupled Tensor Factorization to Model Changing User Preferences for Movie Recommendations

Abstract

In movie recommendation systems, users usually tend to change their preferences over time. Some recent studies suggest that modeling the temporal dynamics of user preferences can improve the quality of recommendations. In this paper, we propose a time-dynamic model of user preferences based on approximately coupled tensor factorization. First, we model the user-item interaction information as a tensor and downweight the user’s historical preferences using an individual exponential decay factor. Second, we extract similarity information from the interaction information as auxiliary information to mitigate the cold-start and data sparsity problems. Then, we use approximately coupled tensor factorization to jointly analyze the obtained data to generate the top-K recommendations. We validate the effectiveness of our proposed method on the MovieLens dataset, and the experimental results show that our method performs better than other competitive methods.