An Enhanced Probabilistic Graphical Model Framework, With Application to Community Identification In Dynamic Networks

Abstract

Real-world complex networks like social and biological networks generally exhibit inhomogeneity, leading to tightly coupled nodes, clusters or communities which serve a crucial functional purpose for the original system. Analyzing these communities in larger networks has become rapidly one among the hot areas in the complex networks. More general inorder to detect these communities in a variety of such complex networks, various strategies have been put forth. However probabilistic graphical model like Non negative matrix factorization (NMF) approach and its derivatives can efficiently address the issue of dynamic community detection in such complex networks, which is impossible often for other dynamic community detection approaches. Thus one may argue that NMF has a lot of flexibility when it comes to dynamic community detection. Unlike the existing approaches, we have formulated a framework comprising of 3 key steps: 1) information dynamics to quantify information propagation among involved nodes, thereby calculating the number of communities in the respective graph; 2) graph-regularization so as to attain precise representation of topology; 3) NMF soft- community-membership vectors so as to allot the nodes to the communities.

We believe strongly that this work can be a valuable resource for research workers who have interest in dynamic community detection, and encourage them to put in more effort so as to increase the versatility of NMF-based dynamic community detection