Research And System Implementation Of Community Discovery Algorithm For Heterogeneous Network Based On Graph Neural Network

With the advent of the era of big data,all walks of life are increasingly closely connected with the Internet industry,followed by an overwhelming number of interactive data,which constitute a virtual social network.The characteristics,structure and development law of these complex networks are studied and analyzed by means of computer technology.In a sense,these data are the true reflection of the real society,which is of great significance to the research and solution of social problems.Community partition in complex networks is the basic research work in the field of complex networks and social computing,which has always been concerned by researchers.Previous research work mainly focused on the community partition based on homogeneous networks,but the community discovery based on heterogeneous networks can explore a more realistic community structure.Through modeling,representation and analysis of multi-dimensional structure,multi-mode information,semantic information,link relationship and other information in heterogeneous networks,community discover in heterogeneous networks is to find a large number of relatively close and stable community structures among nodes.It has important research value for the analysis and mining of network information,public opinion analysis and other research fields.The main work of this paper is as follows.Firstly,we analyze the DBLP literature data,and use the different types of nodes in the literature data and the relationship between them to construct the direct and indirect relationship,so as to form the network structure information of the nodes.Then the web crawler crawls the paper abstracts in the DBLP data set,constructs the content characteristic information of the node,and combines the network structure information and content characteristic information of the node to construct the heterogeneous network data set.Secondly,two neural network models of Graph Convolution Network and Graph Attention Network,are used to analyze the classification effect of graph neural network on heterogeneous network data sets,and the important role of indirect relationship in the research task is further analyzed.In order to solve the problem that K-means clustering algorithm needs to determine the class of community partition and the initial point of community partition in advance,a community partition method is adopted based on graph convolution neural network and agglomerative hierarchical clustering(HAC)algorithm.Compared with the classical community partition algorithm without graph convolution neural network,the results show that the two-stage community partition algorithm based on graph convolution neural network can better aggregate the structure and content of nodes.Because graph convolution neural network has advantages in dealing with multi-dimensional and multi type social network information.Then convolutional neural network combined with HAC algorithm is also better than other methods in heterogeneous network community partition results and efficiency.Finally,this paper designs and implements the display system of the community discovery.The display system is mainly divided into two function module,such as the original network structure display and community detection display.The display module of original network structure mainly visualizes the node distribution and relationship structure of various network data sets.The display module of community detection is a visual display of community partition results on different network datasets using different algorithms of community discovery.