KGFedRS: Knowledge Graph enhanced Federated Recommender System.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Federated recommender systems train models locally to preserve user privacy.
• On-device training often leads to data sparsity, reducing recommendation accuracy.
• Existing methods struggle to balance privacy with the need for rich user data.

Purpose of the Study:

• To propose KGFedRS, a novel federated recommender system enhanced by knowledge graphs (KGs).
• To protect user and KG privacy while mitigating data sparsity through KG auxiliary information.
• To improve the effectiveness and efficiency of federated recommendation.

Main Methods:

• A privacy-preserving framework using a third-party server for encrypted KG and user profile matching.
• A KG-guided implicit interaction subgraph generation module for local client signal learning.
• A local subgraph expansion module to capture explicit high-order collaborative information.

Main Results:

• KGFedRS demonstrates superior performance compared to state-of-the-art federated recommendation methods.
• The system effectively alleviates the data sparsity problem by incorporating KG information.
• Experiments on three public datasets confirm the enhanced effectiveness and efficiency of KGFedRS.

Conclusions:

• KGFedRS successfully integrates knowledge graphs into federated recommendation systems.
• The proposed approach enhances recommendation accuracy and efficiency while maintaining strong privacy guarantees.
• This work offers a promising direction for privacy-preserving and data-efficient recommender systems.