An information-theoretic approach for detecting communities in networks

Detecting communities in a network can be helpful to comprehend its structure and understand its function. The detecting-communities approach, in essence, is a model of classification and belongs to the scope of social methodology. In this paper, we view the community description of a network as a lossy compression of that network’s information, and develop an information-theoretic foundation accordingly for the concept of community in networks. We present an optimization model and its algorithm to identify the communities by finding an optimal compression of the network. We also illustrate the availability of this approach by an artificial example, compare its accuracy and algorithm complexity with the other classical approaches of this field by a series of simulated networks with different parameters, and demonstrate its application in a real-world network. The tests show that the proposed method is a good one for detecting the communities and finding the proper community number in the unweighted and weighted networks.