Depth-Enhanced Contrast Attribute Graph Clustering

Abstract

Attribute graph clustering is an important clustering task. Existing graph clustering methods mostly focus on the structure and attribute information of the graph, often ignoring the potential connections between nodes. Meanwhile, contrastive learning methods mostly rely on predefined graph enhancement methods, which may limit the robustness and reliability of the views. Adaptive contrastive learning methods may not fully consider the graph structure and attribute information when generating views, which may affect clustering performance. To solve this problem, this paper proposes a new attribute graph clustering method that aims to preserve the graph structure and attribute information as much as possible while capturing the relevant information between nodes. First, a deep graph enhancement module is used to obtain a new adjacency matrix under an enhanced view, and contrastive learning is used to optimize the view learning process, ensuring that graph structure and attribute information are considered while fully considering the mutual information between nodes to optimize view generation. Then, a graph neural network (GCN) is used to learn the graph structure and attribute information to obtain graph embeddings. Next, the contrastive learning module ensures that the embeddings preserve the graph structure and attribute information while capturing the information between nodes. Finally, the embedding learning process is supervised by a graph clustering loss to make it applicable to graph clustering tasks. A large number of experiments were conducted on five benchmark datasets, and the results show that the proposed method has good performance for attribute graph clustering.