Intent Enhanced Self-supervised Hypergraph Learning for Session-Based Recommendation

Xiu Susie Fang, Yonggang Wu, Jinhu Lu, Xiaoyu Gu, Guohao Sun, Yong Zhan

Abstract: Session-based recommendation (SBR) aims to capture user intents based on a set of anonymous sessions for recommending the next item. Recent works in SBR often employ graph neural networks (GNNs) to model the transition patterns between items and have made impressive progress. However, the performance is still limited by data sparsity and complex dependency in sessions. Recently, self-supervised learning (SSL) has been applied in recommender systems because of its good ability to mine ground-truth samples from raw data, and great potential in relaxing data sparsity. We note that both sessions and individual items contain implicit user intents, and there is consistency between intents. Therefore, the SSL can be applied to construct self-supervised signals based on the implicit user intents to further alleviate the data sparsity problem in SBR, and thus improve the performance. In this paper, we propose a novel model called Intent Enhanced Self-Supervised Hypergraph Learning for session-based recommendation (ISHGL) to improve the performance. We first model the session sequence data as a global hypergraph to capture complex high-order relationships in sessions. Then, we devise a new contrastive method for self-supervised learning without additional data augmentation and complex positive/negative sample constructions. Extensive experiments on three datasets demonstrate the superiority of our model over the state-of-the-art methods.

External IDs:dblp:conf/pkdd/FangWLGSZ24