Enhanced community detection in social networks using active spectral clustering

Abstract

Detecting community structure is a fundamental problem in social networks analysis. In this paper, an enhanced semi-supervised approach to community detection using active spectral clustering is proposed. This approach incorporates partial background knowledge in the form of pairwise must-link and cannot-link constraints into community detection. It achieves significant performance improvement using fewer constraints by actively selecting the most informative constraints and querying human expert for them. The constraints are incorporated into spectral clustering by adjusting the pairwise similarity matrix accordingly. Experimental results on benchmark synthetic and real world social networks show that this approach significantly outperforms recent semi-supervised algorithms for community detection in terms of the Normalized Mutual Information (NMI) achieved with respects to the percentage of constraints used.