Graph Neural Networks (GNN) have demonstrated significant advantages in processing and analyzing graph-structured data in recent years, and they have been widely applied in various fields such as social network analysis, recommendation systems, knowledge graphs, bioinformatics, chemical molecular graphs, traffic networks, and computer vision. This paper reviews the foundational theories of GNN, including the basic concepts of graphs, representation methods, and key operations such as graph convolution and graph pooling. It thoroughly explores several major GNN models, such as GCN, GAT, Graph Autoencoders, Graph GAN, GIN, and Temporal GNN. Furthermore, the paper analyzes the specific applications of GNN in different fields, common experimental datasets, and evaluation metrics, revealing the current main challenges such as scalability, model interpretability, data sparsity and imbalance, cross-domain graph representation learning, and robustness and security issues. Finally, the paper looks into the future development directions and research hotspots of GNN, aiming to provide comprehensive references for researchers and promote continuous innovation and progress in this field.