Multi-Intent Enhanced Hypergraph Model for Session-Based Recommendation

Session-based recommendation leverages interaction data from anonymous users to suggest items of potential interest. Although existing research typically examines user intent from global and local perspectives, the omission of initial user intent has led to an incomplete understanding of user preferences. Furthermore, during the intent fusion stage, many methods simply sum or concatenate different intents, making it difficult to discriminate the contribution of each intent to the final recommendation. To overcome these limitations, we propose the multi-intent enhanced hypergraph model for session-based recommendation (MEHM). Specifically, we introduce two novel modules: an Initial Intent Learning Module (IILM) and a Multi-Intent Fusion Module (MIFM). In the Initial Intent Learning Module, we propose a first-item-guided strategy that infers the user's initial intent based on the first few items in the session. In the Multi-Intent Fusion Module, we integrate multiple forms of user intent-namely, the initial, global, and current intents-by using global intent as a reference and assigning appropriate weights to both initial and current intents. This design allows for a more precise and holistic representation of user preferences. Extensive experiments on three real-world datasets demonstrate that our method achieves superior performance compared to existing state-of-the-art approaches.