<p>Graph clustering is an important and challenging task in graph data mining. Recently, contrastive learning has been used in this field to learn ‘cluster-preserving’ representations, gradually becoming a mainstream research direction and significantly improving clustering performance. However, most existing state-of-the-art methods heavily rely on graph augmentation techniques. These methods frequently change the way graphs are structured or the details of the nodes when they are augmented, which can cause a loss of meaning and reduce the trustworthiness of the learned representations, making it harder to improve performance further. Moreover, such methods are inherently dependent on the quality of the augmented graphs. To address these issues, this paper proposes a novel augmentation-free method named Structure-Aware Walking: Contrastive Learning Method for Graph Clustering (SWCL-GC). First, we introduce a structure-aware walk (SAW) strategy, which is capable of capturing potential high-order neighbor relationships and applying them to the construction of positive/negative sample pairs. Then, an improved GraphSAGE encoder is employed to encode both structural and jointly attributed information, generating high-quality node representations. Finally, we use contrastive learning to improve the node representations, helping to understand the structure of the data while keeping the semantics of the attributes. Extensive experiments conducted on four widely used graph datasets demonstrate that SWCL-GC outperforms state-of-the-art graph clustering methods in terms of clustering performance.</p>