A Multi-Objective Evolutionary Algorithm with Neighbour NodeCentrality for Community Detection in Complex Networks

Abstract

The field of identifying natural partitions in complex networks has witnessed enormous attention in recent years. However, the mathematical definition to unfold accurate structure for communities is still in need of more investigation. Further, the literature lacks any driving strategy to claim which mathematical definition is to be honored. Our contribution in this paper is threefold. First, we introduce a new score model for the community detection problem. Unlike other state-of-theart models, the proposed model aims to reflect the atom relationships of the nodes rather than that of the communities. The proposed model includes two new contradictory objective functions. The first objective is ruled in favor of tightly grouping the neighbours for each node within one community. The second objective, on the other hand, attempts to disturb the maximum number of external interactions for each node with respect to the node’s internal neighbours. The second contribution comes with designing a new algorithm, a multi-objective evolutionary algorithm based community detection in complex networks (MOEA-CD). The contradictory objectives composing the proposed score model are optimized simultaneously in the MOEA-CD. We also introduce a new local heuristic search operator, namely, Neighbour Node Centrality (NNC) strategy. The strategy is used in the MOEA-CD algorithm to improve its performance. The third contribution comes with proposing a new evaluation technique to validate the quality of the existing and the proposed score functions. The proposed technique is based on a random node migration strategy applied on the ground-truth partition for the network. Empirical evaluation of the proposed model against the state-of-the-art models clarifies that the proposed model is generally more correlated to represent the Normalised Mutual Information (NMI) between a candidate partition and the correct partition. Further, extensive experiments on both real-world and synthetic benchmarks of different complexities demonstrate that the proposed model can reveal more accurate community structures in comparison to the state-of-the-art models. The results also demonstrate the ability of the proposed (NNC) strategy to improve the detection accuracy of all tested models.

A Review of Community Detection Based on Modularity Optimization in Complex Networks

“…In other words, the main problem is to divide a community into partitions (communities), which are characterized by dense connections inside each part but a sparse connection between them. Many community detection methods in social networks and other types of networks are presented; some of them are traditional methods like graph partitioning [1] [2], hierarchical clustering [3] [4], fuzzy clustering [5], partitional clustering such as kmeans clustering with its extensions [6] [7], and spectral clustering [8] [9]. whereas other methods involve the optimization of some quality function, like the modularity function [10] [11], and another technique based on statistical inference that aims to deduce a data set's properties, like techniques based on Bayesian inference [12], block modeling [13], model selection [14], and information theory.…”