Research On Community Discovery Based On Topic Model

In recent years, increasing numbers of researchers turn their attention to complex network. Using the method of community discovery to mine the hidden relationship under the surface of the network is of great importance in both scientific study and commercial application. Since all complex networks have general characters of community structure that all the networks are consisting of a number of communities, using the method of community discovery for detecting their community structures and evolution process is of great importance in the analyzing topology features and dynamic characteristics.It is generally agreed that the main work of community discovery includes two parts:(1) discover the community structure for each time period; (2) discover the community evolution through the relationship between communities. Although domestic and foreign researchers have made intensive studies of cominunity discovery in large-scale network, we find that the accuracy of the high computing efficient methods is rather low while the computing efficient of the high accuracy method is rather low.In order to increase efficiency of community discovery method without accuracy lost, we propose a novel algorithm for community discovery base on LDA topic model. By introducing topic model, we can obtain the abstract subject information from a lot of network data.The results show that our method can improve the efficiency of the network construction related words. In the constructed network, we use clique percolation method to discovery the community structure for each time period. This can accurately describe the network structure and characteristics of community. Then we calculate the community correlation in adjacent time. Through the size comparison of the community correlation value, we discover the relationship between communities. At last, through the analysis of the community content change, we can get community evolution.In this experiment, we construct a word association network about research topics by the topics of DBLP bibliography records, and apply the method above to discovery the community structure and community evolution. Experiment results show our method improves the accuracy of community discovery and reduces the complexity of the algorithm.