Deep Graph Clustering via Mutual Information Maximization and Mixture Model

TL;DR

This paper introduces GMIM, a novel graph clustering method that maximizes mutual information and models node embeddings with a Gaussian mixture, improving community detection accuracy.

Contribution

It proposes a unified framework combining mutual information maximization with Gaussian mixture modeling for graph clustering, which is a novel approach in this domain.

Findings

Outperforms existing methods on real-world datasets

Effectively detects communities with high accuracy

Joint optimization improves embedding quality

Abstract

Attributed graph clustering or community detection which learns to cluster the nodes of a graph is a challenging task in graph analysis. In this paper, we introduce a contrastive learning framework for learning clustering-friendly node embedding. Although graph contrastive learning has shown outstanding performance in self-supervised graph learning, using it for graph clustering is not well explored. We propose Gaussian mixture information maximization (GMIM) which utilizes a mutual information maximization approach for node embedding. Meanwhile, it assumes that the representation space follows a Mixture of Gaussians (MoG) distribution. The clustering part of our objective tries to fit a Gaussian distribution to each community. The node embedding is jointly optimized with the parameters of MoG in a unified framework. Experiments on real-world datasets demonstrate the effectiveness of our method in community detection.