Graph neural networks have shown significant progress in node classification tasks. However, their performance declines when learning from only a few examples per class or when adapting to unseen classes. Meta-learning helps overcome this limitation by enabling models to generalize from limited examples, and adapt to novel classes not encountered during training. In this setting, algorithms train on diverse meta-tasks consisting of a support set containing a few nodes from specific classes and a query set with unseen nodes from those classes, which are used to evaluate their performance based on their ability to adapt to novel data distributions. Current approaches often augment meta-tasks with additional data such as the neighboring nodes of target nodes are used to enrich both support and query sets. However, the structural complexity of graphs introduces challenges in designing effective meta-tasks, as variations in graph structures across tasks can hinder consistent feature representations. To address these challenges, this research explores augmentation strategies in combination with contrastive learning to extract node and class characteristics by identifying instances based on similarity. Evaluation of the effectiveness of this approach through a comprehensive comparison with state-of-the-art methods on benchmark node classification datasets has been demonstrated in this work. This project’s source code is publicly available at  https://github.com/sirajummprince/COS-META .