This paper proposes a novel graph contrastive learning method with adaptive augmentation, named GCA, to improve graph representation learning. The method addresses the challenge of designing effective data augmentation schemes in graph contrastive learning, which is crucial for learning representations that are robust to perturbations. Unlike existing methods that use uniform data augmentation strategies, GCA incorporates adaptive augmentation schemes that preserve intrinsic graph structures and attributes. Specifically, on the topology level, GCA uses node centrality measures to highlight important connective structures by removing edges with lower importance. On the node attribute level, GCA corrupts node features by adding noise to unimportant features, forcing the model to learn semantic information. The method is evaluated on five real-world datasets, and results show that GCA consistently outperforms existing state-of-the-art baselines and even surpasses some supervised methods. The proposed framework is effective in learning robust graph representations by adaptively augmenting the graph structure and attributes.This paper proposes a novel graph contrastive learning method with adaptive augmentation, named GCA, to improve graph representation learning. The method addresses the challenge of designing effective data augmentation schemes in graph contrastive learning, which is crucial for learning representations that are robust to perturbations. Unlike existing methods that use uniform data augmentation strategies, GCA incorporates adaptive augmentation schemes that preserve intrinsic graph structures and attributes. Specifically, on the topology level, GCA uses node centrality measures to highlight important connective structures by removing edges with lower importance. On the node attribute level, GCA corrupts node features by adding noise to unimportant features, forcing the model to learn semantic information. The method is evaluated on five real-world datasets, and results show that GCA consistently outperforms existing state-of-the-art baselines and even surpasses some supervised methods. The proposed framework is effective in learning robust graph representations by adaptively augmenting the graph structure and attributes.