Paper Notes: On Community Detection in Real-world Networks and the Importance of Degree Assortativit

Paper title: On Community Detection in Real-world Networks and the Importance of Degree Assortativit

Author: Marek Ciglan, Michal Laclavik, Kjetil Norvag

Conference: KDD 2013, Chicago, Illinois, USA

Year: 2013

This paper studies the relation between thedetected communities of three near-linear community detection algorithms andthe ground truth community. The results show that on two networks (Orkut andDBPedia) these community detection methods fail to deliver good communitystructure.  A further investigationreveals that the majority of the nodes have been signed to a few of the largestclusters on the two networks.

Based on the above observation, the author studies a measure named degree assortativity coefficient, which denotes a tendency of nodes to be connected with other nodes of similar degree. The results show that networks with assortative community structure have important parts of communities composed of edges connecting nodes with similar degrees. Then the author proposes to weight the edges in a way that lower the weight of the edges connecting disassortative nodes. Two edge-weighting functions are presented to lower the influence of the edges connecting low and high degree nodes and the experiments prove that the two functions can significantly improve the results of community detection methods on networks with assotative community structure.