Abstract
Temporal graph clustering is a complex task that involves discoveringmeaningful structures in dynamic graphs where relationships and entities changeover time. Existing methods typically require centralized data collection,which poses significant privacy and communication challenges. In this work, weintroduce a novel Federated Temporal Graph Clustering (FTGC) framework thatenables decentralized training of graph neural networks (GNNs) across multipleclients, ensuring data privacy throughout the process. Our approachincorporates a temporal aggregation mechanism to effectively capture theevolution of graph structures over time and a federated optimization strategyto collaboratively learn high-quality clustering representations. By preservingdata privacy and reducing communication overhead, our framework achievescompetitive performance on temporal graph datasets, making it a promisingsolution for privacy-sensitive, real-world applications involving dynamic data.