Personal Interest Attention Graph Neural Networks for Session-Based Recommendation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Session-based recommendation systems predict user clicks using historical and current sessions.
• Existing methods struggle with complex item transitions and dynamic user interests.
• Current approaches often use fixed session vectors, neglecting real-time user preferences.

Purpose of the Study:

• To propose a novel session-based recommendation model that accurately captures item transitions and user interests.
• To address limitations in existing methods regarding dynamic user preferences and complex item interactions.
• To enhance the accuracy and personalization of recommendations in e-commerce and app environments.

Main Methods:

• Developed a personalized interest attention graph neural network (PIA-GNN).
• Utilized personalized graph convolutional networks (PGNN) to model item transitions.
• Incorporated an interest-aware mechanism and self-attention for adaptive user interest and long-term dependency capture.
• Trained the model using cross-entropy loss.

Main Results:

• PIA-GNN demonstrated superior performance compared to existing personalized session-aware recommendation methods.
• The model effectively captures complex item transitions and adapts to users' evolving interests.
• Experiments on two real-world datasets validated the proposed approach's effectiveness.

Conclusions:

• PIA-GNN offers a significant advancement in session-based recommendation accuracy.
• The model's ability to handle dynamic user interests and item interactions is key to its success.
• This approach provides a more personalized and effective recommendation experience.