Multilayer graph contrastive clustering network

Abstract

Listen

Translate article

Multilayer graphs have received significant research attention in numerous areas beacause of their high utility in modeling interdependent systems. However, clustering of the multilayer graph, in which multiple networks divide the graph nodes into categories or communities, is still at a nascent stage. Existing graph-clustering methods are often restricted to exploiting the multiview attributes or multiple networks and ignore more complex and richer network frameworks. Thus, we propose a generic and an effective autoencoder framework for multilayer graph-clustering called a multilayer graph contrastive clustering network (MGCCN). The MGCCN consists of three modules: (1) attention mechanism that is applied to better capture the relevance between nodes and their neighbors for better node embeddings. (2) A contrastive fusion strategy that efficiently explores the consistent information in different networks. (3) A self-supervised component that iteratively strengthens the node embedding and clustering. Extensive experiments on different types of real-world graph data indicate that our proposed method outperforms other state-of-the-art techniques.