Detection Of Generalized Semantic Communities By Robustly Combining Network Topology And Node Content

Community detection is an important task in the field of complex network research.Using of network topology information and content information for community detection and community interpretation has become a new development trend in the field of complex network analysis.However,there are several problems in the current related research work.First,most methods consider the definition of a community as a collection of points with tightly connected internal nodes and sparse connections between groups.Therefore,traditional community detection method based on this concept cannot detect other types of community structures that exist in the real world,that is,generalized community structures.Second,in the real network,there may be multiple topics in a community.Many methods often assume that a community corresponds to only one topic,and the resulting community interpretation is inaccurate.Finally,most methods use words to represent the topic of community interpretation.However,words often have multiple meanings,so only using words to represent topics alone may cause ambiguity and is not easy to understand.To solve the above problems,we propose two novel community detection and interpretation methods which combine network topology information and node content information.We model the real network through the Bayesian probability framework and design an efficient variational expectation maximization algorithm to learn the model parameters.By characterizing the membership of the community,the topic clustering of the nodes,and the description of the relationship between the two,the method we propose can use multiple topics to explain the results of the community.We further use the information at the emotional level to characterize the distribution of emotions under each theme in the community,thereby obtaining the "topic-emotional" double-level community interpretation result.We compare our models with several state-of-the-art algorithms in artificial networks and real networks to verify the accuracy of our methods.We also provide case studies to show the ability of our two progressive algorithms in semantic interpretability,respectively.