Recently social relationship among users has been exploited to improve the recommendation performance. The intuition behind most of these work is social homophily such that users are more similar to their neighbors. Attention mechanism or attention network from deep learning has been a popular component employed by recommendation models. However, how to attentively learn the influence between users remains pretty much open in the existing social recommendation models. In this paper, we propose a social recommendation model MAS, Multi-head Attentive Social Recommendation. The key to MAS is a multi-head attention network which can distinguish the impact of users' friends when predicting users' preference on different items. When compared to the state-of-the-art baseline methods on three real-world datasets, our method achieves the best performance.
展开▼
