Detecting communities in complex networks-A discrete hybrid evolutionary approach

Evolving densely connected communities of nodes in the real life complex networks is a computationally extensive (NP hard) problem. Nature based evolutionary heuristic algorithms provide an effective solution to such kind of problems. However, very few evolutionary approaches have been tested on this domain with most of them applying real time operators. Incorporating discrete behaviour in the evolutionary process can further lead to improvisation in the overall efficiency of the applied algorithm. This paper proposes a discrete adaption of the TL-GSO community detection (CD) algorithm for faster convergence of the optimization function in comparison to the existing variants of Group Search Optimization (GSO) and TL-GSO. The approach mixes the exploration strategies of Teachers Learners (I-TLBO) and GSO algorithms to detect communities in complex networks. It modifies the optical search of GSO to step search and real time crossover to single point crossover. The modifications result in minimizing the parameters to be externally set. Optimized search space reduces the runtime and evolve communities in an unsupervised manner. Experimental results on real and synthetic networks show that the proposed algorithm converges faster and evolves to accurate communities with high fitness as compared to varied state of the art CD algorithms. © 2016 Informa UK Limited.