Title: Efficiently mining community structures in weighted social networks

Authors: Hédia Zardi; Lotfi Ben Romdhane; Zahia Guessoum

Addresses: LIP6, University of Paris-VI, France; MARS Research Group, University of Monastir, Tunisia ' MARS Research Group, SDM Team, ISITCom, University of Sousse, Tunisia ' LIP6, MAS Team, University of Paris-VI, France

Abstract: In the literature, there are several models for detecting communities in social networks. In Zardi and Romdhane (2013), we presented a robust method, called maximum equilibrium purity (MEP), in which we defined a new function that qualifies a network partition into communities, and we presented an algorithm that optimises this function. We proved that, unlike modularity-based models, MEP does not suffer from the 'resolution limit' problem. However, MEP operates only on unweighted networks; i.e., networks where all connections are considered equally. Hence, strengths of social ties between network nodes are ignored. Unfortunately, this assumption may not hold in several real-world networks where tie strengths play a major role. In this paper, we present the maximum weighted equilibrium purity algorithm (MWEP), the extension of MEP to weighted networks. Like the original model, the extended model is proved to circumvent the 'resolution limit' problem encountered in community detection. In addition, we have applied our model to real-world and synthetic social networks and experimental results are more than encouraging.

Keywords: social networks; weighted graphs; community detection; objective function; data mining; community structures; online communities; virtual communities; web based communities; resolution limit.

DOI: 10.1504/IJDMMM.2016.075969

International Journal of Data Mining, Modelling and Management, 2016 Vol.8 No.1, pp.32 - 61

Published online: 20 Apr 2016 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article