Graph contrastive learning methods have emerged as a prominent research focus on deep graph clustering, with data augmentation serving as a pivotal element of contrastive learning. Existing methods typically generate augmented views randomly. Although these methods have achieved some success, randomly generating augmented views heavily relies on predefined strategies, which limits the flexibility of data augmentation and reduces the semantic consistency of augmented views. To address these challenges, we propose a novel approach, namely Learnable Deep Contrastive Graph Clustering with Centrality Guidance (GUIDE). Specifically, we introduce a stage-focused attention encoder that adaptively adjusts the attention weights of different nodes to generate more reliable representations. To ensure that the sampled subgraph maximizes local information while preserving global key nodes, we design a centrality guidance module to enrich information both at the node and edge levels. In addition to the contrastive objective, we also introduce a deep clustering objective, which is jointly optimized to improve the performance of the model in the clustering task. Extensive experiments conducted on five datasets demonstrate the effectiveness of the GUIDE method.