In the cutting-edge exploration of the recommendation system field, Graph Contrastive Learning (GCL) has attracted much attention for its outstanding contribution to data augmentation capabilities, becoming a new research hotspot. However, existing models encounter bottlenecks in maintaining the integrity of graph structure semantics and enhancing robustness against noise interference. To break through this limitation, this paper innovatively proposed a graph contrastive learning model based on Principal Component Analysis (PCA) information supplementation. This model, for the first time, integrates PCA into the contrastive learning framework, accurately identifying and extracting key features from the graph, significantly enhancing the semantic richness of node embeddings. On this basis, by constructing high-quality positive and negative sample pairs, contrastive learning is conducted from multiple dimensions such as graph structure and node semantics, deeply mining the high-order features of the graph, and thereby obtaining a complete semantic representation of the entire graph. Furthermore, a structure and collaborative view encoding learning module is constructed, expanding the horizon of graph contrastive learning through cross-view collaborative contrastive learning, enriching the semantics of the prediction vector, and enhancing the model’s ability to resist noise interference. Through comparative experiments on three standard datasets, the outcomes demonstrate that the performance of the model introduced in this study exceeds that of various established baseline models, thereby confirming the efficacy of this approach.