In this paper, we propose to integrate contrastive learning into a point cloud representation framework to enhance the learning of 3D point cloud features. Three-dimensional (3D) point cloud provide rich geometric detail that is essential for applications such as autonomous driving, robotics, and virtual reality. Dynamic Graph Convolutional Neural Networks (DGCNNs) demonstrate strong performance in capturing local geometric structures through dynamic graph construction and edge convolutions. However, effectively learning latent spaces that capture both local and global semantic information remains challenging. To address this, we incorporate the normalized temperature-scaled cross-entropy (NT-Xent) loss into the framework, which encourages embeddings of semantically similar shapes to cluster closer in the latent space while pushing dissimilar shapes further apart. Experiments on the ModelNet10 dataset show that our method achieves higher classification accuracy compared to the baseline DGCNN. Qualitative analyses, including shape correspondence evaluations, further indicate that our approach captures both local and global features more effectively than the baseline. These results highlight the potential of combining graph-based feature extraction with contrastive learning, offering a promising direction for robust and generalizable 3D point cloud representation learning.