Consistency-Regularized Multi-Stage Joint-Perception Graph Fuzzy Clustering Algorithm

Traditional clustering methods struggle to simultaneously account for both local features and global structure within data. Existing deep graph-based clustering approaches also exhibit limitations in extracting hierarchical structural information from data, resulting in constrained clustering performance on complex datasets. To address the aforementioned issues, this paper proposes a Consistency-Regularized Multi-Stage Joint-Perception Graph Fuzzy Clustering Algorithm (C-MSGFC). This approach employs a parallel architecture combining a Graph Autoencoder (GAE) with a Graph Attention Autoencoder (GATE) to achieve joint perception of global structure and local topological features. It enhances the discriminative power and robustness of the embedded representations through a Gated Dynamic Heterogeneous Fusion module (GDHF). Moreover, a multi-stage clustering strategy with consistency constraints is introduced to effectively extract hierarchical structural information, thereby enhancing the stability and generalization capability of clustering results. Experimental results demonstrate that C-MSGFC achieves superior clustering performance compared to existing SOTA methods across multiple public datasets. Data and code can be found at: https://github.com/yyy123-UX/C-MSGFC .