Contrastive Learning in Adjacency Matrix Based Deep Graph Clustering

Deep graph clustering is an important part of cluster analysis and achieves fuzzy classification in unsupervised scenarios by learning a pseudo-label for each node. Deep graph clustering is often based on the adjacency matrix, and the learning of node representation is achieved by operating on the entire adjacency matrix. Since real-world graph data are often full of noise or have missing information, it is difficult in most cases to directly obtain an effective supervisory signal. Therefore, in unsupervised scenarios, contrastive learning is able to mine information from the data to generate more contrast signals for the model, thus enhancing the performance of the model. Here we introduce several key technologies of graph clustering and contrastive learning, supplemented by an introduction to commonly used datasets for deep graph clustering for analysis. Finally, we discuss the current challenges and future prospects for applications of graph contrastive clustering.