A New Multi-objective Evolution Model for Community Detection in Multi-layer Networks

Abstract

In reality, many complex network systems can be abstracted to community detection in multi-layer networks, such as social relationships networks across multiple platforms. The composite community structure in multi-layer networks should be able to comprehensively reflect and describe the community structure of all layers. At present, most community detection algorithms mainly focus on the single layer networks, while those in multi-layer networks are still at the initial stage. In order to detect community structures in multi-layer networks, a new multi-objective evolution model is proposed in this paper. This model introduces the concept of modularity in different decision domains and the method of local search to iteratively optimize each layer of a network. Taking NSGA-II as the benchmark algorithm, the proposed multi-objective evolution model is applied to optimize the genetic operation and optimal solution selection strategies. The new algorithm is denoted as MulNSGA-II. The MulNSGA-II algorithm adopts the locus-based representation strategy, and integrates the genetic operation and local search. In addition, different optimal solution selection strategies are used to determine the optimal composite community structure. Experiments are carried out in real and synthetic networks, and results demonstrate the performance and effectiveness of the proposed model in multi-layer networks.