Rule-guided Skip-GCN in neural latent information diffusion network for social recommendation

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Existing graph neural network (GNN) methods for social recommendation often oversimplify diverse social influences.
• These methods neglect crucial latent item-item correlations, hindering accurate modeling of user preferences.

Purpose of the Study:

• To propose DiffRSG, a novel framework addressing limitations in current social recommendation approaches.
• To enhance recommendation accuracy by integrating neural latent information diffusion with explicit rule-guided reasoning.

Main Methods:

• Developed DiffRSG, a framework employing a rule-guided graph convolutional network (GCN) with skip connections (Skip-GCN).
• Utilized Skip-GCN to capture heterogeneous social interactions and uncover implicit item dependencies via explicit rules.
• Incorporated a specialized prediction layer for precise rating estimation.

Main Results:

• DiffRSG significantly outperformed eleven baseline models on Yelp and Flickr datasets.
• Achieved an average relative improvement of over 15% in hit ratio (HR) and normalized discounted cumulative gain (NDCG) compared to the strongest baseline.
• Demonstrated superior performance on HR@10 and NDCG@10 metrics.

Conclusions:

• DiffRSG effectively differentiates social tie strengths and captures latent item correlations within a unified framework.
• The proposed method offers a significant advancement in social recommendation systems.
• Validated the framework's effectiveness through extensive experimental evaluations.