An entropy-based social network community detecting method and its application to scientometrics

Abstract

Community structure is one of the important properties of social networks in general and in particular the citation networks in the field of scientometrics. A majority of existing methods are not proper for detecting communities in a directed network, and thus hinders their applications in the citation networks. In this paper, we provide a novel method which not only overcomes the above mentioned disability, but also has a relative low algorithm time complexity which facilitates the application in large scale networks. We use the concept of Shannon entropy to measure a network's information and then consider the process of detecting communities as a process of information loss. Based on this idea, we develop an optimal model to depict the process of detecting communities and further introduce the principle of dynamic programming to solve the model. A simulation test is also designed to examine the model's accuracy in discovering the community structure and identifying the optimal community number. Finally, we apply our method in a citation network from the journal Scientometrics and then provide several insights on promising research topics through the detected communities by our method.