Research On Community Discovery Method Based On Node Structure And Content Similarity In Complex Networks

In the current multidisciplinary research hotspot: The research of complex networks has drawn more and more attention,which has great significance and far-reaching influence in the fields of computer science,physics,biology,social science and management science.Community discovery is a very important research direction in complex network research.To strengthen the understanding of the structure of community detection in complex networks,function analysis and behavior prediction has important theoretical significance and practical value,has been widely used in the identification of terrorist organization,recommended prediction,search engine and network public opinion monitoring and many other fields.How to quickly,accurately find the complex network community(community discovery)is still a key problem.The existing community discovery method is mainly based on the network topology,it is difficult to deal with large-scale network.In recent years,though there is also a community discovery method that not only improves the quality of community discovery by introducing text content,but also applies to large-scale network clustering problem.However,these algorithms have higher algorithm dimension,so it is difficult to accurately depict the content representation of nodes,and the computation is large.Therefore,research and development of high clustering accuracy,efficient algorithm execution,and suitable for large-scale complex network of community discovery methods become the research hotspot.Based on this,the main research object of this paper is the community discovery method in complex networks.The specific work includes the following two aspects: 1)Complex networks can be abstracted as an evolutive graph with complex connections and structures.In view of the analysis and division of social identities in the past more intuitive graphs,we propose a Gaussian Mixture Model(GMM)that can effectively handle the content of the nodes,and better integrate the similarity of the node structure and content information on the complex network.The complex network the graph is sampled to construct a simplified graph,and finally the graph is clustered.Because this method can adopt different parameters to fit the data points in the Gaussian mixture model,it has better scalability.2)Since the time complexity of graph clustering is relatively high,and the high dimension complex network data basically can be converted into a matrix form for processing,so we propose a modular Non-negative Matrix Factorization(NMF)model that combines the similarity of node structure and content information to match fitting the original complex network to get the most accurate network of nodes,and finally using K-means clustering algorithm for clustering.The method is concise and easy to operate,and has the interpretability of the attribute relationship of the community,so it has more accurate representation of the relationship between the nodes and the community.In this paper,we propose two new method of community discovery that combines the structure and content similarity of nodes,which is verified experimentally in real networks.The performance and accuracy of dimension computation are obviously better than those of previous methods,thus further improving the accuracy and efficiency of community discovery.