A Survey of Graph Federation Learning for Data Privacy Security Scenarios

Abstract In recent years, the research on Graph Neural Networks(GNNs) and Federated Learning(FL) has become more and more mature. However, these studies rely heavily on training data. With the improvement of user data privacy awareness and the emphasis on data security at the national level, the training data has become relatively large. difficulty. Moreover, in practical application scenarios, most of the data is unstructured, making it very important to adopt GNNs to learn representations. This paper will give a comprehensive overview of the fusion learning algorithm of GNNs and FL under the framework of distributed machine learning from the perspective of data privacy and security. First, an overview of the current challenges of graph federated learning(GFL); then, an in-depth analysis of GFL methods, and an extended exploration of the topology of graphs; then, focus on several graph-structured federated learning privacy calculations, communication mechanisms, etc. Secondly, it makes a comparative analysis of the common methods of GFL, experimental data sets and experimental evaluations; finally, this paper looks forward to the research hotspots of GFL in combination with privacy and security.