Modeling Transition Matrix for a Collaborative Rating Prediction Recommendation System via Nonnegative Tensor Decomposition

We investigate the problem of user preferences changing over time for a collaborative rating prediction recommendation system. The rating data of a time-aware collaborative recommendation can be represented as a 3rd order tensor, each of which denotes users, items and time. We mainly focus on establishing a user preference transfer model according to the time latent factor matrice, which is computed by tensor decomposition. Subsequently, a novel nonnegative tensor decomposition with temporal similarity (TS-NTD) is put forward to predict rating values. The proposed temporal similarity measure succeeds in describing various changes in user preferences. We evaluate the TS-NTD on two real world datasets, and experimental results show that our model has higher accuracy than state-of-the-art models.