User preference interaction fusion and swap attention graph neural network for recommender system

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Recommender systems are crucial for personalized user experiences.
• Knowledge graphs enhance recommendations by providing rich relational information.
• Existing methods struggle to leverage fine-grained knowledge graph details and user importance.

Purpose of the Study:

• To propose a novel knowledge-graph-based graph neural network (PIFSA-GNN) for improved recommendation performance.
• To address limitations in current methods regarding fine-grained knowledge graph utilization and user-entity importance.
• To enhance the aggregation of neighboring entities by considering user preferences.

Main Methods:

• Developed PIFSA-GNN, a knowledge-graph-based graph neural network.
• Incorporated user preference interaction fusion to integrate auxiliary user information.
• Implemented user preference swap attention for improved entity weight calculation and aggregation.

Main Results:

• PIFSA-GNN demonstrated superior performance on movie, restaurant, and music datasets.
• Significant improvements observed in AUC, F1-score, Hit@1, Hit@5, and Hit@10 metrics compared to baseline methods.
• Outperformed the best baseline by up to 2.6% in AUC and 7.2% in F1 on the restaurant dataset.

Conclusions:

• PIFSA-GNN effectively utilizes fine-grained knowledge graph information and user preferences.
• The proposed method enhances the accuracy and effectiveness of recommender systems.
• The approach offers a promising direction for future research in knowledge-graph-enhanced recommendations.