Community detection (CD) is a powerful way to extract meaningful information from networks such as political election networks, biological networks, social networks, technological networks. This study proposes a modified discrete version of Coot bird natural life model (COOT) optimization algorithm to solve CD problem in the networks. The basic COOT method is based on the different collective behaviors of the birds of the coot family. These collective actions of coots are regular and irregular movements on the water surface. The position update rule of the basic COOT method does not provide a balance between exploitation and exploration ability for the problem addressed in this study. Therefore, a new update mechanism is integrated into the basic COOT method to extend the local and global search tendencies of the basic COOT method. In the proposed COOT method (for short MCOOT), in order to create a new position for the current coot individual, first the original update mechanism of COOT method is carried out; then, the proposed update mechanism is executed. Three important modifications have been made in the new update mechanism: (1) Some dimensions of the current coot individual are randomly selected in the range of 1 to the dimension size of the problem; (2) the selected dimensions of the coot individual are updated according to the proposed update rule; (3) a genetic mutation operator is executed on the current coot position according to a mutation probability to improve the exploration ability. Furthermore, in the proposed MCOOT method, the continuous values of the current coot positions are converted to discrete values, because the CD problem is a discrete problem. Based on these modifications, in order to analyze and validate the effectiveness of the proposed MCOOT, it is applied on ten different small-sized or large-sized network problems. Finally, the experimental results of MCOOT method are compared with those of some state-of-the-art optimization methods in terms of solution quality and time evaluation. According to the experiments of our study, the proposed algorithm is obtained the best results for all community detection problems used in this study when compared with 22 other algorithms. As a result, the proposed method achieves superior or comparable performance in terms of solution quality and robustness according to the general results. Therefore, the proposed method can be much more competitive, especially for discrete problems.
Read full abstract