A Label-Based Nature Heuristic Algorithm for Dynamic Community Detection

Abstract

The evolving patterns of the real-world can be tracked and captured by the dynamic network community structure. Some existing methods such as the multi-objective particle swarm optimization (MOPSO) use the evolutionary clustering model to detect the dynamic network community. However, the MOPSO has defects that are undesirable premature convergence and insufficient diversity of particles due to a high selection pressure. Therefore, a label-based heuristic algorithm based on the evolutionary clustering model is proposed for overcoming those shortcomings. The label propagation algorithm is adopted to initialize community structure and restrict the condition of the mutation process. The operations of crossover and mutation are used to increase the diversity of solutions and maintain the quality of community structure simultaneously. Experimental results demonstrate that the proposed method is effective and outperforms other methods in synthetic and real-world datasets.