Informative Anchor-Enhanced Heterogeneous Global Graph Neural Networks for Personalized Session-Based Recommendation

Ronghua Lin, Luyao Teng, Feiyi Tang, Hao Zhong, Chengzhe Yuan, Chengjie Mao

Abstract: Due to the anonymity of user sessions, most existing session-based recommender systems (SBRSs) cannot effectively learn user features, leading to failure to make personalized recommendations. Besides, these SBRSs may neglect some similar items with common features if they are long-distance in the session graphs or global graphs. In this paper, we propose a novel SBRS based on heterogeneous graph neural network, which can effectively learn user and item embeddings for personalized recommendations. Furthermore, we find out the user and item informative anchors in the heterogeneous graph and propagate their features in the same type of nodes, which can help to explore those long-distance but similar items. We conduct extensive experiments on three real-world datasets and the experimental results demonstrate the effectiveness of our proposed method.