MIGAN: Mutual-Interaction Graph Attention Network for Collaborative Filtering.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommender systems are prevalent on web platforms.
• Network representation learning is a key technique for building efficient recommender systems.
• Learning mutual node influence in networks is crucial for capturing collaborative signals in user-item interactions.

Purpose of the Study:

• To develop a novel algorithm for effective network representation learning in recommender systems.
• To address the challenge of learning mutual node influence to better understand user decisions.
• To improve prediction accuracy and recommendation efficiency.

Main Methods:

• Developed Mutual Interaction Graph Attention Network (MIGAN), a new algorithm.
• Utilized self-supervised representation learning.
• Applied the algorithm to a large-scale bipartite graph (BGNN).

Main Results:

• MIGAN demonstrates superior performance compared to existing baseline methods.
• Experimental results show significant improvements in prediction accuracy.
• The algorithm achieves enhanced recommendation efficiency.

Conclusions:

• MIGAN effectively captures complex user-item interactions through mutual node influence.
• The proposed method offers a promising approach for advancing network representation learning in recommender systems.
• MIGAN provides a robust solution for improving the performance of web platform recommender systems.