Using network topology and semantic contents to find topic-related communities is a new trend in the field of community detection for attributed networks with words content.Through networked content analysis,we find that,in most cases,the semantic structure of networked content is multiplex.That is,there exists a background topic to reflect the background information of the whole complex network,and two-level topics(i.e.general and specialized topics)to respectively represent common and specific interests of each community.However,the existing community detection methods ignore such a multiplex semantics structure and take all words used to describe node semantics from an identical perspective.This indiscriminate use of words leads to natural defects in depicting networked content in which the multiplex semantics are not be fully utilized.To address this issue,we progressively propose two novel Bayesian probabilistic frameworks to model real networks,and devise efficient variational expectationmaximization algorithm for model inference.By distinguishing the background topic and two-level topics of words,finally,our model not only can find community structures more accurately,but also utilizes background topic words to reflect the background information of all communities and describe each community using both specialized topic(to denote their particular interests)and general topic(to denote their shared features with similar communities).The superiority of our algorithms for finding community structures in networks are further demonstrated in comparison with eight state-of-the-art algorithms on eight real-world networks,respectively.In experiments,we also provide case studies to show the ability of our two progressive algorithms in multiplex semantic interpretability,respectively. |