Abstract:In recent years, graph clustering has emerged as one of the most challenging problems in the field of deep learning. With the increasing complexity of real-world networks, such as social and biological networks, more and more effective methods are needed to organize and understand these structures. Various cognition-based techniques have been explored for classifying nodes within these graphs, and graph convolution networks (GCNs) have attracted great interest. GCNs, a deep semi-supervised learning approach, provide a powerful framework for learning node representations by utilizing both local and global graph information. By iteratively aggregating information from neighboring nodes, GCNs effectively capture the intricate relationships and dependencies within complex networks. We introduce a novel deep unsupervised learning scheme built upon the foundation of GCN architecture. The key contributions are outlined below. First, the entire architecture is trained with three unsupervised learning losses. The first loss focuses on kernelized features that use node attributes to reflect the information extracted from the data. The second loss leverages spectral smoothness that uses connections between nodes to capture global cluster structure. The third loss is based on graph reconstruction that introduces additional regularization of the representation of nodes by the output of the model. Second, the spectral smoothing loss involves an adaptive approach using an additional graph matrix associated with the node representations. This adaptive integration of additional structural information increases the learning efficiency during the training phase. The adaptive fused graph used for spectral smoothing loss incorporates structural insights derived from both data features and deep node representations. To assess the performance of our approach, we conducted extensive experimental evaluations on four benchmark datasets widely used in the field of graph clustering. These datasets were carefully selected to cover diverse domains and varying degrees of complexity, ensuring a comprehensive evaluation of our method's efficacy. Our results showcase the remarkable performance of our unsupervised GCN across multiple metrics, surpassing other state-of-the-art graph neural network techniques in terms of clustering accuracy, purity, and other relevant measures. Notably, our method consistently outperforms competing approaches across the used datasets, demonstrating its versatility and effectiveness in handling various real-world scenarios.

SEGCN: Structural Enhancement Graph Clustering Network

Graph Clustering Network with Structure Embedding Enhanced.

Synergistic Deep Graph Clustering Network

Structure-enhanced Contrastive Learning for Graph Clustering

CaEGCN: Cross-Attention Fusion based Enhanced Graph Convolutional Network for Clustering

Self-Supervised Contrastive Graph Clustering Network via Structural Information Fusion

Adaptive Graph Convolutional Clustering Network with Optimal Probabilistic Graph

Representation Learning Method of Graph Convolutional Network Based on Structure Enhancement

Structure Enhanced Deep Clustering Network Via a Weighted Neighbourhood Auto-Encoder

Adaptive Unsupervised Graph Convolution Network for Data Clustering with Graph Reconstruction

Deep Fusion Clustering Network With Reliable Structure Preservation

Spectral Embedding Network for Attributed Graph Clustering.

Attention-driven Graph Clustering Network.

Multi-scale Graph Clustering Network

Attribute Graph Clustering Based on Self-Supervised Spectral Embedding Network

Dynamic Graph Attention-Guided Graph Clustering with Entropy Minimization Self-Supervision

SHCN: Self-supervised General Hypergraph Clustering Network

Multi-scale graph attention subspace clustering network

Towards attributed graph clustering using enhanced graph and reconstructed graph structure

Deep Structural Enhanced Network for Document Clustering

Enhancing Graph Collaborative Filtering Via Neighborhood Structure Embedding