Exploiting Multi-Facet User Intent Information for Session-Based Recommendation

Session-based recommendation (SBR) aims to predict users' next actions based on anonymous and short behavior sequences. Recently, graph neural networks (GNN) have been widely applied in related research to capture the complex transition relationships between items. However, most methods achieve marginal improvements by relying on complex GNN architecture designs and ignoring the leverage of diverse user interaction intent information. To fully exploit user intent information to improve recommendation performance without making the model overly complex, we propose a Multi-facet User Intent Information Exploiting Network (MUIIEN). Specifically, MUIIEN recommend items mainly via two novel modules we designed: (i) the Parallel-Gate Pivot GNN (PGP-GNN) module, which learns complex transition patterns between items by using a simple architecture with multiple parallel gates and a transition-controlling pivot; and (ii) the Multi-facet User Intent Mixture (MUIM) module, which exploits three facets of user intent information with dual-facet user intent queries and an attention-based intent mixture mechanism. Experiments on three real-world datasets show that the MUIIEN model is superior to the state-of-the-art SBR methods.