Time-aware tensor factorization for temporal recommendation

In recent years, temporal recommendation, which recommends items to users with considering temporal information has attracted widespread attention. How to capture and combine the time-varying user behavior distributions and the time-varying user behavior transition patterns is challenging. To address these challenges, we propose a Time-Aware Tensor Factorization for Temporal Recommendation (TATF4TRec). First, the personalized Markov transition tensors are applied to represent the users’ temporal behaviors. Then a tensor factorization method is proposed to capture the time-varying patterns of these tensors. Furthermore, the model linearly combines the time-varying patterns of user behavior and predicts the recommended results at a given time. Extensive experiments on five datasets demonstrate that TATF4TRec outperforms the state-of-the-art baselines significantly.