A Multi-objective Evolutionary Algorithm Based on Multi-layer Network Reduction for Community Detection.

Community detection is an important method to reveal the characteristics of complex systems, which usually requires the system to meet the conditions of close connections within communities but sparse connections between communities. In view of this, community detection has been proven to be an NP-hard problem. Multi-objective evolutionary algorithm (MOEA) is an indispensable aspect of multi-layer network community detection. However, most MOEA-based multi-layer network detection algorithms only take the acquired prior information as the network preprocessing method and ignore its full utilization in optimization, resulting in the accuracy of network partition cannot be guaranteed. To this end, this paper proposes a multi-objective community detection algorithm based on multi-layer network reduction (MOEA-MR). Specifically, we use the non-negative matrix factorization method to generate the consistent prior information layer of multi-layer network. Based on this, a network reduction strategy based on node degree is constructed to recursively reduce the size of the prior information network. In addition, in the evolution process, we consider using the multi-layer network similarity to correct the mis-divided nodes in the local reduction community. Compared with other advanced community detection algorithms, the experimental results on the real-world and synthetic multi-layer networks proved the superiority of MOEA-MR.