A Fast Parallel Genetic Algorithm Based Approach for Community Detection in Large Networks

For large scale networks, under various formulations, the community detection problem is known to be NP-hard. Several genetic algorithm based community detection methods have been reported in the literature so far. But most of them exhibit a very slow convergence and hence not scalable. To achieve faster convergence, in this paper, we propose a parallel genetic algorithm for community detection based on diameter of the communities. To the best of our knowledge, so far, ours is the first work where the diameter of the community has been considered as a key parameter for community detection. This new concept expedites the convergence of the proposed genetic algorithm significantly without sacrificing the quality of the solution. Implementation of the algorithm on GP-GPU platform finds good quality community structures with near optimal modularity value with significantly less number of generations. Experiments show that it achieves 3 to 5 times speedup on synthetically created LFR (Lancichinetti-Fortunato-Radicchi) benchmark networks compared to the existing sequential genetic algorithm. Simulation on well-known real-world benchmark networks with up to 58 K nodes and 428 K edges results speedup of 4.27 at maximum.