Graph-Augmented Multi-Level Representation Learning for Session-based Recommendation

Abstract

Session-based recommendation has received extensive attention in recent years due to its excellent performance in recommending for anonymous users. However, most existing studies ignore the intention revealed by several consecutive items in a session and do not consider similar transitional relations across sessions. To this end, we propose a graph-augmented multi-level representation learning method for session-based recommendation. Specifically, we construct multi-level session graphs for each session to capture fine-grained user preferences. In addition, we utilize subgraphs of global transition graph to augment the representation of each session. Contrastive learning is adopted to maximize the consistency of the representation of the same session under different subgraphs. Experimental results on two real-world datasets prove that the proposed method achieves excellent performance compared to several competitive baselines.