Intention Enhanced Dual Heterogeneous Graph Attention Network for Sequential Recommendation

Abstract

Sequential recommendation plays a vital role in many web applications, aiming to predict users’ next actions based on their historical sequential behaviors. Efficiently learning the features of items and understanding user’s intentions are pivotal for sequential recommendation. However, due to the diversity of items and the sparseness of user’s interaction with items, it’s challenging to accurately learn the features of items through sparse interaction data. In addition, users usually have shifting intentions when interacting with items, which makes it difficult to understand users’ intentions. To this end, we propose a novel intention enhanced dual heterogeneous graph attention network (IE-DHGAT) for sequential recommendation. Specifically, we construct an extensible heterogeneous graph, which contains items and items’ various attributes, and we design a dual graph attention network to learn the features of items via explicitly incorporating item’s various attribute information into item embeddings. Further, we propose an intention enhanced attention layer to efficiently capture users’ shifting intentions through computing the correlation between items and discriminating different intention areas in users’ interaction sequences. We conduct extensive experiments on three real-world datasets and the results demonstrate that our proposed approach achieves better performance than the state-of-the-art methods.