Research Article
Enhancing Session-Based Recommendation with Multi-granularity User Interest-Aware Graph Neural Networks
@INPROCEEDINGS{10.1007/978-3-031-54531-3_16, author={Cairong Yan and Yiwei Zhang and Xiangyang Feng and Yanglan Gan}, title={Enhancing Session-Based Recommendation with Multi-granularity User Interest-Aware Graph Neural Networks}, proceedings={Collaborative Computing: Networking, Applications and Worksharing. 19th EAI International Conference, CollaborateCom 2023, Corfu Island, Greece, October 4-6, 2023, Proceedings, Part III}, proceedings_a={COLLABORATECOM PART 3}, year={2024}, month={2}, keywords={Recommender system Session-based recommendation Graph neural network Self-supervised learning}, doi={10.1007/978-3-031-54531-3_16} }- Cairong Yan
Yiwei Zhang
Xiangyang Feng
Yanglan Gan
Year: 2024
Enhancing Session-Based Recommendation with Multi-granularity User Interest-Aware Graph Neural Networks
COLLABORATECOM PART 3
Springer
DOI: 10.1007/978-3-031-54531-3_16
Abstract
Session-based recommendation aims at predicting the next interaction based on short-term behaviors within an anonymous session. Conventional session-based recommendation methods primarily focus on studying the sequential relationships of items in a session while often failing to adequately consider the impact of user interest on the next interaction item. This paper proposes theMulti-granularityUserInterest-awareGraphNeuralNetworks (MUI-GNN) model, which leverages item attributes and global context information to capture users’ multi-granularity interest. Specifically, in addition to capturing the sequential information within sessions, our model incorporates individual and group interest of users at item and global granularity, respectively, enabling more accurate item representations. In MUI-GNN, a session graph utilizes the sequential relationships between different interactions to infer the scenario of the session. An item graph explores individual user interest by searching items with similar attributes, while a global graph mines similar behavior patterns between different sessions to uncover group interest among users. We apply contrastive learning to reduce noise interference during the graph construction process and help the model obtain more contextual information. Extensive experiments conducted on three real-world datasets have demonstrated that the proposed MUI-GNN outperforms state-of-the-art session-based recommendation models.
