Tensor-Based Federated Graph Clustering

A graph is a collection of nodes and edges.Clustering analysis is an important technology in the field of data mining.It can explore the internal relationship of data from multiple angles and find the similarity of data.It can be used for community detection,network partition,user behavior prediction and other tasks.With the development of social science and technology,data appear in more and more complex scenes,and it is no longer in a single independent form.They are interrelated and influence each other,and more graph data appear in the field of data analysis.The research of graph is a topic that people are gradually interested in in in the process of exploring the clustering algorithm.Graph clustering can classify the network,and divide the data into different clusters by knowing the characteristics of nodes and referring to the structure of graph network.Due to the existence of massive multi-source heterogeneous data in the real world,the traditional graph nodes and graph structure expressed by numerical or vector can not satisfy the description of complex data gradually.This paper introduces the concept of tensor to more fully represent the sample nodes,and also uses graph structure to more comprehensively explore the relationship between nodes from multiple perspectives.When multiple participants want to cooperate in graph clustering,it is necessary to consider the efficiency and privacy protection.At present,the distributed graph processing tasks are mainly focused on sparse graph processing to improve efficiency.For privacy issues,there is only a single graph task node anonymization privacy protection technology,and there is no suitable multi-party cooperation privacy protection scheme.The federated learning framework is used to complete the task of multi-party cooperation graph clustering,which meets the needs of privacy data mining and analysis.The concepts of tensor and federated learning are used to optimize the graph clustering algorithm to obtain a multi-user secure tensor graph clustering model.Firstly,the deep attention embedding graph clustering is extended to the high-dimensional tensor space,and the adjacency tensor is used to represent the relationship between the same group of nodes in different relationships,while the node tensor can describe the different characteristics of complex samples in different relationships,which provides more sufficient and interpretable clustering results for real-world clustering tasks.Then simulate multiple clients,join the federal learning framework,make multiple graph clustering tasks with the same distribution exchange training gradient in the learning process,in order to get a better effect,more universal graph clustering model.Finally,by comparing the tensor based federated graph clustering algorithm with other graph clustering algorithms which only use a single eigenvalue or feature matrix,traditional clustering algorithm and one-way computing clustering algorithm,the experimental results show that this algorithm has better performance and efficiency,and can provide privacy protection for multi-party secure graph clustering.