Self-supervised global context graph neural network for session-based recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Session-based recommendation (SBR) relies on limited user interaction data within short timeframes.
• Existing methods struggle with noisy interactions and capturing complex, high-order item relationships across sessions.

Purpose of the Study:

• To propose a novel graph neural network model for SBR that effectively captures high-order item transition relations.
• To enhance model robustness and representation learning through a self-supervised contrastive learning module.

Main Methods:

• Developed a self-supervised global context graph neural network (SGC-GNN) utilizing virtual context vectors.
• Implemented a contrastive self-supervised learning (SSL) module as an auxiliary task for robust representation learning.
• Modeled high-order transition relations between items across all sessions, connecting virtual context vectors to all items within a session.

Main Results:

• The proposed SGC-GNN model demonstrated superior performance compared to state-of-the-art methods on three benchmark datasets.
• Validated the effectiveness of virtual context vectors in capturing item relations beyond adjacent items.
• Confirmed the positive impact of the self-supervised learning module on model robustness and overall SBR task performance.

Conclusions:

• The SGC-GNN effectively models complex, high-order item transition relations crucial for SBR.
• The integration of virtual context vectors and contrastive self-supervised learning significantly improves recommendation accuracy and robustness.
• This approach offers a promising direction for advancing session-based recommendation systems.