Identifying overlapping communities in networks using evolutionary method

Abstract

Community structure is a typical property of real-world networks, and has been recognized as a key to understand the dynamics of the networked systems. In most of the networks overwhelming nodes apparently live in a community while there often exists a few nodes straddling several communities. Hence, an ideal algorithm for community detection is that which can identify the overlapping communities in these networks. We present an evolutionary method for detecting overlapping community structure in the network. To represent an overlapping division of a network, we develop an encoding scheme composed of two segments, the first one represents a disjoint partition and the second one represents an extension of the partition that allows of multiple memberships. We give two measures for the informativeness of a node, and present a coevolutionary scheme between two segments over the population for solving the overlapping partition of the network. Experimental results show this method can give a better solution to a network. It is also revealed that a best overlapping partition of the network might not be rooted from a best disjoint partition.