TL-GSO: - A hybrid approach to mine communities from social networks

Mining communities of densely connected nodes with high modularity from social network graphs is an NP hard problem. Researchers have applied various soft computing evolutionary optimization techniques to evolve communities by optimizing modularity. This paper proposes an efficient hybrid evolutionary algorithm TL-GSO which incorporates the exploration and exploitation capabilities of two recently proposed nature inspired algorithms named Improved-Teacher´s Learner´s Based Optimization (I-TLBO) and Group Search Optimization (GSO) algorithms. The hybridization has resulted in fast convergence to accurate communities as compared to the results of applying GSO with single producer. TL-GSO adapts the strategy of teacher´s learners by dividing the search space into small groups and allocating one teacher/producer to each group. Multiple producers improves the fitness of the population more efficiently by independently evolving small scrounger groups in a generation. The producer scrounger (PS) search strategy of GSO is then applied in groups to evolve them locally. Further scroungers behavior is also modified to make them perform single point crossover instead of real coded crossover in order to improve their fitness. The algorithm was tested on four real world datasets and results were compared with evolutionary as well as non-evolutionary state of the art algorithms. The results shows the effectiveness of proposed TL-GSO algorithm over many state of the art algorithms.