Self-Attention Based Time-Rating-Aware Context Recommender System

Abstract

The sequential recommendation can predict the user’s next behavior according to the user’s historical interaction sequence. To better capture users’ preferences, some sequential recommendation models propose time-aware attention networks to capture users’ long-term and short-term intentions. However, although these models have achieved good results, they ignore the influence of users on the rating information of items. We believe that in the sequential recommendation, the user’s displayed feedback (rating) on an item reflects the user’s preference for the item, which directly affects the user’s choice of the next item to a certain extent. In different periods of sequential recommendation, the user’s rating of the item reflects the change in the user’s preference. In this paper, we separately model the time interval of items in the user’s interaction sequence and the ratings of the items in the interaction sequence to obtain temporal context and rating context, respectively. Finally, we exploit the self-attention mechanism to capture the impact of temporal context and rating context on users’ preferences to predict items that users would click next. Experiments on three public benchmark datasets show that our proposed model (SATRAC) outperforms several state-of-the-art methods. The Hit@10 value of the SATRAC model on the three datasets (Movies-1M, Amazon-Movies, Amazon-CDs) increased by 0.73%, 2.73%, and 1.36%, and the NDCG@10 value increased by 5.90%, 3.47%, and 4.59%, respectively.