Proximity, Communities, and Attributes in Social Network Visualisation

Abstract

The identification of groups in social networks drawn as graphs is an important task for social scientists who wish to know how a population divides with respect to relationships or attributes. Community detection algorithms identify communities (groups) in social networks by finding clusters in the graph: that is, sets of people (nodes) where the relationships (edges) between them are more numerous than their relationships with other nodes. This approach to determining communities is naturally based on the underlying structure of the network, rather than on attributes associated with nodes. In this paper, we report on an experiment that (a) compares the effectiveness of several force-directed graph layout algorithms for visually identifying communities, and (b) investigates their usefulness when group membership is based not on structure, but on attributes associated with the people in the network. We find algorithms that clearly separate communities with large distances to be most effective, while using colour to represent community membership is more successful than reliance on structural layout.