This research introduces a novel data-centric approach to tackle heterogeneity problems in federated graph learning (FGL) while preserving privacy.

• Leverages large language models to address non-IID data distributions across clients
• Enables collaborative model training without sharing sensitive raw graph data
• Enhances both convergence and performance of federated graphs
• Prioritizes security by only transmitting model parameters between clients

This innovation matters for security teams as it provides a framework for organizations to collaborate on graph learning tasks while maintaining strict data privacy compliance.

Data-centric Federated Graph Learning with Large Language Models