Community Detective And Influence Analysis Based On Deep Learning

With the rapid development of Internet and mobile network,social network is also growing vigorously.At the same time,the identification of community structure in social networks is of great significance to aware of the network’s evolution function and topological structure.Nowadays,many community detective researcher has been successfully applied to recommendation system,individuation product promotion,protein function prediction,sentiment analysis and other fields.However,with the amplify of the network scale,the traditional community detective research has some limitations and high computing costs when dealing with large-scale network.Especially when using scriptures clustering algorithm to process high-dimensional feature data,the results of traditional community detective research have insufficient accuracy.In this paper,we use deep learning methods to conduct community detective on social network,and analyzes the node influence within community structure to find the more important nodes.The work of this paper is summarized as follows:(1)This paper proposes dynamic node2 vec algorithm(d-node2vec)and graph convolution network fuses modularity.In this paper,we integrate the attribute features of nodes into the random walk of node2 vec,so that the network structure and the attributes of nodes are considered comprehensively in the random walk process,which is more suitable for social network’s structure,and generate higher accuracy network representation.In the graph convolution network,we incorporate the concept of modularity into the Laplace matrix to learn the community structure of the network.Because the D-node2 vec algorithm pay more attentions to local random walk of nodes,the resulting representation pays more attention to the local features of nodes,and the graph convolution network that fuses modularity is more concerned with the global features of nodes.So,we fuse the two network representation methods,and get the network representation method of this paper,experiments show that the fusion method can enhance the performance of network representation.(2)This paper designs community classification model based on global and local features fusion.In this paper,we improved the capsule graph neural network by introducing the network representation learning method we proposed,and designed a community classification model based on global and local features Fusion,which improves the accuracy of community classification.Then,on the foundation of community classification model based on global and local features Fusion,we introduce the random path of D-node2 vec to improve the connection between capsules,and increase the local connection between the lower and higher capsules,and propose semi-supervised community detective model based on Classification.This model can perform community classification tasks and community discovery tasks.The experimental results show that the model’s accuracy is slightly higher than community classification model based on global and local features Fusion.In the community detective task,the model has achieved better effects than traditional methods and deep learning methods.(3)In this paper,after getting the obvious community structure,we uses Page Rank algorithm combined community structure to analyze the importance of nodes within community.We consider the importance of node attributes in the analysis of node influence,and limit the importance of overlapping nodes,proposing Page Rank algorithm based on community structure.Compared to Page Rank,its complexity can be reduced,also avoid the theme drift of Page Rank.Besides,it can consider the abnormal nodes in the network,and optimize the influence ranking of nodes.