Graph neural networks for preference social recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Social recommendation systems leverage social network information to enhance user recommendations.
• Existing graph neural network (GNN) approaches face challenges with social network data, including misaligned preferences between connected users and information redundancy in user representations.

Purpose of the Study:

• To propose a novel graph neural network (GNN) framework for preference-based social recommendation.
• To address the limitations of current GNN applications in social recommendation, specifically preference misalignment and information redundancy.

Main Methods:

• Introduced a 'friend influence indicator' to create a Social Preference Network, reflecting user preference similarity.
• Employed distinct GNNs for the Social Preference Network and user-item interaction network to prevent information redundancy.
• Utilized two loss functions to maintain connection relationships and differentiate between positive and negative samples.

Main Results:

• The proposed Preference Social Recommendation (PSR) model demonstrates effectiveness in recommendation tasks.
• PSR is lightweight and shows particular strength in addressing cold-start recommendation problems.
• Experimental results validate the model's ability to improve recommendation performance.

Conclusions:

• The proposed graph neural network approach effectively integrates social preference information for improved recommendations.
• PSR offers a lightweight and efficient solution, particularly beneficial for cold-start scenarios in recommendation systems.