GAFGCN: A graph augmented fusion network for enhanced deep clustering in attributed graphs

Attributed graph clustering aims to partition nodes into cohesive groups based on structural and attribute similarities. We propose a novel Graph Augmented Fusion Clustering Network (GAFGCN), which introduces an enhanced graph construction strategy and captures multi-scale features from local-to-global perspectives. GAFGCN consists of four key modules: (1) an Information-enhanced Encoder Module (IEM) for initial node representation, (2) a Graph Augmentation and Information Extraction Module (GAIE) for constructing enhanced graphs and capturing high-order information, (3) a Parallel Information Learning Module (PILM) for multi-scale feature extraction, and (4) a Self-Adaptive Fusion Learning Module (SFLM) for adaptively fusing node representations and generating clustering results. Extensive experiments conducted on nine datasets demonstrate the effectiveness and superiority of our proposed method over the state-of-the-art approaches.