Dynamic Community Detection Based on a Label-Based Swarm Intelligence

Abstract

The dynamic network tails after the development of the real-world that is essential for particle applications such as traffic flow analyses and social network analyses. The requirement of maximizing the quality of the community structure at current time step and minimizing the difference of the community structure between two successive time steps synchronously brings serious challenges to the dynamic community detection. Some existing approaches (i.e., the multi-objective particle swarm optimization, named as DYNMOPSO) utilize the swarm intelligence pattern to solve such a community detection problem in dynamic networks. Nevertheless, the DYNMOPSO has the deficiency of the undesirable prematurity constringency and monotonicity of particles because of the high choice stress. Thus, a label-based swarm intelligence on the basis of the evolutionary clustering framework is presented for overcoming those disadvantages. The label propagation approach initializes the labels of particles and is used for escaping the prematurity constringency. The crossover and mutation methods are introduced to improve the variety of particles and retain preferable the community structure synchronously. Experiments in synthetical and real networks prove that our algorithm is valid and exceeds state-of-the-art approaches.