Temporal Topic Model Based Community Detection

With the development of Internet technology,various types of network structure are becoming more and more complex,as it is difficult for users to find useful information directly,the research on complex network has been widely concerned by scholars from domestic and foreign.Complex networks have general characteristics of community structure.Community detection is one of the important technologies in the complex network analysis which can detect the structure of community effectively and help people analyze and understand the topology structure in the network.Meanwhile,the structure of complex network is changing with the time constantly in the real life,tracing the evolution of community can help researchers understand the dynamic trend of network.In this paper,we analyze some current community detection algorithms and find some shortages in the traditional community detection method.In order to find community structure from a large-scale network more accurately,we propose a community detection method based on temporal topic model which combining the structure of network with the content attributes of network node.Firstly,we introduce some traditional topic models and community detection method,meanwhile,we analyze and compare their advantages and disadvantages.Secondly,cited papers and year of papers are added to the LDA model,and reference the thought of dynamic topic model to partitioning time slice,we propose Temporal Citation Topic Model.Then combining the importance of users in the network,the result of author topic distribution for each time slice is more accurate.Thirdly,Overcoming the shortages of traditional label propagation algorithm which without considering the node's content attributes and the randomness of label update process,traditional label propagation algorithm is improved,and the topic weight label propagation algorithm is proposed in order to detect different topic community structure in each time slice.Then according to the change of topic and community structure over time,we analyze their evolution process.In the background of social network,starting from the user's impact on the network,the whole network is divided into different communities according to the community detection method based on temporal topic model.In this paper,Firstly,We construct author cooperation network by the cooperation relationship between paper's authors in the DBLP dataset.Then extracting the information of paper titles?year and authors,using the community detection method based on temporal topic model to detect community structure in different time slice,analyzing the evolving of community.Finally,the topic model's perplexity and community detection's modularity are employed to compare and evaluate the experiments.Experiment results show our method improves the accuracy of community detection effectively on the premise of guarantee time complexity.