Context-embedded hypergraph attention network and self-attention for session recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Session-based recommendation systems face challenges in modeling user intent from limited historical data.
• Existing methods often overlook item consistency and context adaptation within user sessions.
• Deep learning approaches primarily focus on sequential or pairwise item relationships.

Purpose of the Study:

• To propose a novel session-based recommendation model, C-HAN, that addresses limitations in current approaches.
• To enhance user intention modeling by capturing inherent item consistency and sequential dependencies.
• To improve context adaptation in session-based recommendation systems.

Main Methods:

• Developed C-HAN, a model featuring parallel modules: a context-embedded hypergraph attention network and self-attention.
• Utilized hypergraph attention to incorporate diverse interaction contexts, boosting contextual awareness.
• Employed a soft-attention mechanism to integrate sequential and consistency information for session representation.

Main Results:

• C-HAN demonstrated superior performance over state-of-the-art methods on three real-world datasets.
• Achieved average improvements of 6.55% in Precision@K, 5.91% in Recall@K, and 6.17% in MRR.
• Validated the effectiveness of integrating item consistency, sequential dependence, and contextual information.

Conclusions:

• The proposed C-HAN model effectively models user intention in short-term sequences.
• Incorporating context and item consistency significantly enhances session-based recommendation performance.
• C-HAN offers a promising advancement for personalized recommendation systems.