Federated graph-level clustering network with adaptive knowledge compensation

In recent years, the rapid proliferation of graph data in practical applications has underscored the growing significance of distributed graph computing. The emergence of a federated graph-level clustering framework injects strong momentum into the field. However, the personalized knowledge disparity across clients often leads to consensus failure, thereby compromising the performance of the global model. To address this challenge, we propose a novel federated graph-level framework to effectively mitigate the issue of personalized knowledge discrepancies. Specifically, in the client, we develop the Local Knowledge Enhancement (LKE) strategy, which extracts reliable knowledge from diverse sources and refines it through global prototype correction to ensure the generation of high-quality, clustering-oriented representations. While in the server, we develop a Global Prototype Alignment (GPA) mechanism that constructs potential affinity relationships to adaptive divide communities, enabling global optimization of knowledge alignment across clients under the premise of semantic similarity. Experimental evaluations on multiple benchmark datasets demonstrate that the proposed framework outperforms existing state-of-the-art methods, achieving superior global consistency while maintaining high levels of personalized performance.