<p>Attributed graph clustering is an unsupervised learning method designed for graph-structured data. The primary objective of this method is to integrate node attribute information and topological structure to cluster nodes exhibiting similar features and connectivity relationships. However, noise in the original graph structure and node features impedes the model’s ability to extract key information, ultimately diminishing clustering accuracy. Moreover, relying solely on a single encoder configuration limits the model’s capacity to effectively capture global features, thereby restricting its clustering performance. To address these challenges, this paper proposes a deep self-supervised global attention model (DSGAC). The model primarily extracts global topological information by leveraging a global attention mechanism. Additionally, it introduces a Laplace smoothing filter to denoise and smooth node attribute features. Simultaneously, the model employs the generated soft clustering assignments to consistently guide its iterative updates, thereby improving clustering performance. Extensive experiments have been conducted to evaluate the proposed model’s performance. The experimental results demonstrate that DSGAC outperforms 16 state-of-the-art clustering models, including GLAC-GCN, DCP-DEC, and ADIIN, achieving improvements of 1.86%, 2.60%, and 1.97%, respectively. These enhancements are particularly notable in terms of accuracy when evaluated on the Cora, Pubmed, and ACM datasets. The findings of this study validate the proposed model’s effectiveness and superiority in clustering tasks.</p>