HISIR: End2End Hierarchical Intent-Based Sequential Recommendation

Sequential recommendation is pivotal for modern web services, yet accurately capturing the multifaceted nature of user needs remains challenging. Existing intent-aware methods attempt to model user intent, but they often suffer from critical limitations: (1) a Monolithic Interest Representation that fails to capture concurrent user interests, and (2) a Flattened Intent Modeling approach that overlooks the inherent hierarchy of user goals. To address these issues, we propose HISIR, an End2End Hierarchical Intent-based Sequential Recommendation framework. It first employs a Multi-Expert Interest Extraction Layer to disentangle a user’s diverse interests. Subsequently, a Hierarchical Intent Modeling Layer maps these interests into a structured, multi-level representation. The entire framework is optimized by a Dual-Decoupling Loss Optimization Layer to ensure the independence and logical consistency of the learned representations. Extensive experiments on four public benchmarks demonstrate that HISIR consistently and significantly outperforms state-of-the-art baselines. The code repository is available at https://github.com/JessieJiangyan/HISIR .