An Multi-Aspect Attentional Model To Capture Multistratal Influence In Social Group

Abstract

The recommendation system is an effective means to handle “information overload”, and it is also one of the most common applications of big data technology. As an important sub-field in recommendation algorithms, social recommendation effectively enhances performance via using social relationships in online social network platforms. However, there are still two problems in the existing social recommendation models: Firstly, user's preference for item has multiple aspects of information, and its social neighbors only match or influence the target user's preferences in some aspects; Secondly, different friends have varied and dynamic influence on user at target item. To solve the problems, we propose a novel social recommendation model: SMAAM (Social-aware Multi-Aspect Attentional Model). We designed a multi-layer attention mechanism to capture the different levels of influence of social neighbors on users and their multiple aspects of views of items, and then select the most valuable friends and aspect information to model user preferences. The experimental results on two different public datasets explain that our model has more stable and excellent performance than some previously published models. In addition, we have also done some visualization of experimental process data to verify the effectiveness and correctness of the model on learning process and design ideas.