Dynamic overlapping community detection algorithm based on topic segmentation

Abstract

The study of dynamic overlapping semantic communities is an important research content for people to analyze the development pattern of events, monitor the trend of public opinion and commercial personalized recommendation. However, existing dynamic community detection algorithms ignore the influence of attributes on community structure, which not only affects the accuracy of community structure but also fragments the relationship between community and topic, and the similarity of topics within the community is not high. Therefore, this paper proposes a dynamic community detection algorithm based on topic division (DOCTS), which uses node attribute features to make topics, combines the network structures of adjacent moments to get weighted subnetworks, then uses Louvain algorithm to detect weighted subnetworks to get topic sub-communities, and finally, merges and optimizes the communities that will have high similarity to get overlapping topic communities. It is verified in real dataset Facebook that the algorithm can effectively detect communities in dynamic networks and also analyze the community structure based on topic evolution.