<p>Graph Neural Networks (GNNs) have demonstrated strong capabilities in representation learning for graph data across various applications. However, many real-world graphs are often noisily labeled, which could significantly degrade the performance of GNNs. The reason is that noisy labels hinder effective supervision during model training, leading GNN-based models to memorize incorrect samples, thus ultimately degrading performance on downstream tasks. Moreover, approaches for addressing label noise in images and text often rely on a large number of labeled samples and the assumption of data independence, making them less effective in handling label sparsity and structural dependencies in graphs. To address this challenge, we propose a novel approach Robust Noise-Resistant Graph Contrastive Learning (RNRGCL) based on graph contrastive learning to mitigate the impact of label noise and reduce model reliance on label information. Considering both label noise and data sparsity, we introduce an unsupervised contrastive learning paradigm that captures invariant graph information while preserving local structural information. Specifically, we design the local information contrastive loss, which constructs positive and negative samples through data augmentation and higher-order neighborhood information, facilitating the learning of more informative node representations. To further enhance representation learning, we incorporate the re-aggregation head to optimize node embeddings, then introduce the consistency regularization term to supervise model training. Extensive experiments on four real-world datasets validate the effectiveness of our proposed RNRGCL for the node classification task on graphs under different label noise ratios.</p>