Research On Attribute Graph Clustering Algorithms Based On Network Embedding And Leader Recognition

With the rapid development of Internet technology,social network,as an important form of complex network,is more and more closely related to human life.Community structure is the most common and important feature in complex networks.In community structure,nodes within the same community are closely connected,while nodes between different communities are sparsely connected.The community discovery method to reveal the network community structure is of great significance to analyze the network topology and predict the behavior information of nodes in the network.The community structure in complex networks has attracted more and more attention and research in academia and industry.The attribute network has the highest fit with the real-life network,but it is still a big challenge to perfectly integrate topology information and attribute information of the node in,and find the tightly coupled community structure from the attribute network.In this paper,the method of combining topology information of the node with attribute information of the node in the clustering process of attribute graph is studied deeply;this paper use network embedding model to fuse topology information and attribute information of the node;this paper put forward the generalized definition of “Leader node”,and the “Leader node” is applied to AGCNELI(Attributed Graph Clustering-based on Network Embedding and Leader Identification)algorithm.The specific research contents of this paper are as follows:1)Due to the network topology information and node properties for independent heterogeneous data form,data information is not complete and contains noise,so we firstly build corresponding Laplacian matrix of topology information and node properties,and then find the corresponding eigenvectors of the matrix.the embedded space and achieve the goal of removing noise.2)The embedding spaces corresponding to topological information and node information of the attribute graph exist in an independent form,so it is not guaranteed that the degree of correlation between the two fused embedding Spaces can reach the maximum.In this paper,the canonical correlation analysis method is adopted to maximize the degree of association of the embedded space,so as to ensure that the community structure finally mined has the characteristics of compact intra-community node connection and sparse inter-community node connection.3)A large number of comparative experiments were used to verify the proposed method.Experimental results verify the effectiveness of the community discovery method of attribute graph clustering based on network embedding and leader recognition.