Enhancing Session-Based Recommendation With Multi-Interest Hyperbolic Representation Networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Session-based recommendation (SBR) predicts user actions within a single session, distinct from traditional methods relying on user history.
• Current SBR models often use graph networks in Euclidean space, which can fail to capture complex session structures and user interest diversity.

Purpose of the Study:

• To propose a novel Multi-Interest Hyperbolic Representation Network (MIHRN) for enhanced session-based recommendation.
• To address limitations in modeling hierarchical structures and diverse user interests within short, complex user sessions.

Main Methods:

• Employs hyperbolic geometry to model intricate high-order spatial structures and sequential relationships among items.
• Utilizes a hyperbolic hypergraph neural network to capture high-order relationships and local clustering within sessions.
• Incorporates a multi-aspect interest representation module to model the diversity of user interests.

Main Results:

• The proposed MIHRN achieved significant performance improvements across three real-world datasets.
• Demonstrated notable gains under the P@10 metric, with improvements of 23.81%, 14.81%, and 36.84%.

Conclusions:

• MIHRN effectively models complex session dynamics and diverse user interests using hyperbolic geometry.
• The approach offers a promising direction for advancing the accuracy and capability of session-based recommendation systems.