A Semi-supervised Multi-objective Evolutionary Algorithm for Multi-layer Network Community Detection

In the real world, many complex systems can be abstracted as multi-layer networks. Recently, community detection for multi-layer networks plays a vital role in multi-relationship complex system analysis, thus gradually gaining popularity especially in the optimization algorithms. The multi-objective optimization (MOOP) methods attract attention owing to the flexibility in solving community detection problems. Nevertheless, most of the MOOP methods pay little attention to the prior information, which cannot ensure the high-level accuracy and robustness against networks with complicated community structures. To address the problem, this paper proposes a semi-supervised multi-objective evolutionary algorithm for multi-layer community detection (SS-MOML). The SS-MOML mainly consists of two steps: First, it extracts the prior information from the network. Second, based on the prior information, the prior layer is constructed by creating virtual connections and the high-quality initial population is generated. And then the optimization process begins, in which the genetic operation based on the prior information is committed to guiding the evolutionary direction of chromosomes. Some extensive experiments are implemented and the results prove that the proposed SS-MOML stands out in accuracy and robustness than 7 state-of-the-art multi-layer community detection algorithms.