FedSA-GCL: A Semi-Asynchronous Federated Graph Learning Framework with Personalized Aggregation and Cluster-Aware Broadcasting

FedSA-GCL is a semi-asynchronous federated graph learning framework that addresses inefficiencies in existing synchronous methods by incorporating a ClusterCast mechanism, which leverages inter-client label distribution divergence and graph topological characteristics. Evaluated on real-world graph datasets, it outperforms 10 baseline methods, achieving an average improvement of 1.9% with the Louvain algorithm and 3.0% with Metis. This framework is significant for practitioners as it enhances robustness and efficiency in federated graph learning, making it more applicable to real-world scenarios.