Federated Clique Percolation for Overlapping Community Detection on Attributed Networks

Community detection is a popular research topic in complex network analysis, which can be applied in many real-world scenarios such as disease prediction. With the increase of people's awareness of privacy protection, more and more laws enforce the protection of sensitive information while transferring data. The anonymization-based community detection methods have to sacrifice accuracy for privacy protection. In this paper, we first propose a standalone clique percolation algorithm to detect overlapping communities on attributed networks. A clique similarity metric is designed to percolate cliques accurately. Second, we develop a federated clique percolation algorithm to detect overlapping communities on distributed attributed networks. Perturbation strategy and homomorphic encryption are used to protect network privacy. The experiments on real-world and artificial datasets demonstrate that the federated clique percolation algorithm achieves identical results to the standalone ones and realizes higher accuracy than the simple distributed ones without federating learning.