Title: Multi-level fuzzy overlapping community detection algorithm for weighted networks

Authors: Ali Choumane; Ali Harkous

Addresses: Faculty of Sciences, Lebanese University, LaRIFA Lab, Nabatieh, Lebanon ' Faculty of Sciences, Lebanese University, LaRIFA Lab, Nabatieh, Lebanon

Abstract: To understand the hierarchical structure of large networks we propose a multi-level community detection algorithm. Each level consists of a set of fuzzy overlapping communities where a node can belong to multiple communities associated with belonging membership degrees. Our algorithm works for weighted and unweighted networks as well. For each level of hierarchy, it identifies the centres of potential communities and iteratively expands them to form the final communities. A new vector representation of the nodes is proposed and used in the centre's expansion process and in the computation of the belonging degrees. Communities detected at a given level serve as super nodes while identifying the overlapping communities of a higher level. Experiments achieved on the well-known LFR benchmark showed high performance which is measured by the normalised mutual information (NMI). Unlike existing methods, our algorithm shows good and stable performance when varying the number of communities of the overlapping nodes.

Keywords: network analysis; community detection; overlapping communities; hierarchical communities; fuzzy overlapping; neighbourhood overlap.

DOI: 10.1504/IJDS.2019.105259

International Journal of Data Science, 2019 Vol.4 No.4, pp.351 - 375

Received: 24 Jul 2019
Accepted: 08 Nov 2019
Published online: 22 Feb 2020 *

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