Sociological-Theory-Based Multitopic Self-Supervised Recommendation.

Area of Science:

• Computer Science
• Social Network Analysis
• Artificial Intelligence

Background:

• Social relationships provide valuable data for recommendation systems.
• Existing methods suffer from data sparsity and noise, and overlook complex user interactions.
• There's a need for advanced methods to leverage social network intricacies for enhanced recommendations.

Purpose of the Study:

• To introduce a novel sociological-theory-based multitopic self-supervised recommendation method (SMSR).
• To address limitations of current social recommendation techniques, including sparsity, noise, and overlooked interaction details.
• To improve recommendation accuracy by integrating user attitude and dynamic topic modeling.

Main Methods:

• Developed the Social-theory-based Multitopic Self-supervised Recommendation (SMSR) method.
• Integrated user attitude into social relationship construction and employed dynamic routing for topic identification.
• Utilized motifs to reveal higher-order user relations and combined light graph convolutional networks with balance theory.
• Devised four self-supervised signals from heterogeneous graph analysis and social theory.

Main Results:

• SMSR effectively mitigates social noise and improves recommendation accuracy.
• The method successfully identifies and categorizes topics, revealing complex user relations.
• Demonstrated superior performance over state-of-the-art methods in extensive experiments on public datasets.

Conclusions:

• SMSR offers a significant advancement in social recommendation systems.
• The integration of sociological theories and self-supervised learning enhances the exploitation of social graph information.
• The proposed method provides a robust solution for overcoming challenges in social recommendation.