Multi-attention Item Recommendation Model Based on Social Relations

Incorporating social relations in recommendation provides a promising way to alleviate problems of sparsity and cold start in collaborative filtering methods. However, most existing methods do not yet take into account social relations in a relative complete way. Besides the differences between preferences of friends, connection strength and expertise differences of users on a given item also have impacts on the spread of preference between friends. In this paper, we propose a social-aware recommendation model named Multi-Attention Item Recommendation model based on Social relations (MAIRS) which allows us to select more informative friends from the perspectives of their preferences, connection strengths, and expertise on items by their own respective attention models. And then, the three attention models are fused together by utilizing an aggregation function. We compare our method with state-of-the-art models on three real-world datasets: Delicious, Ciao and Epinions. The experimental results show that our method consistently outperforms state-of-the-art models in terms of several ranking metrics.