Research On Complex Network Community Detection Based On Network Embedding And Consensus Community Inserting

Identifying communities is an essential but challenging task in complex network research.Community discovery,as an important method in the field of complex network research,has attracted the attention of researchers in recent decades.With the introduction of modularity,the measurement function of community structure,the community discovery algorithm based on modularity have been unprecedentedly developed.This kind of methods model community discovery as an optimization problem,and then finds the optimal solution with respect to a pre-defined objective function.But most of single-objective optimization methods face some essential difficulties in complex networks.Compared with these,the multi-objective optimization algorithms have the advantages of strong search capability,efficient fitness evaluation and fast convergence speed,while often encounter some problems such as massive search space and low efficiency.In order to solve these problems,the network embedding method is introduced into the multi-objective particle swarm optimization algorithm to map nodes to the low-dimensional space and mine the latent and valuable structure information of the complex network.By designing a consensus-based update strategy,the evolution direction of particles have been constrained and adjusted effectively and tradeoff between two variants of the objective function(KKM and RC)has been obtained.Our method reduces the search space and improves the convergence speed and accuracy effectively.The comparison performance on two synthetic network datasets and 11 real-world network datasets demonstrate that our method not only has good diversity,but also improves the convergence of the algorithm,and makes the distribution of optimal solutions more uniform and wider.We also propose an intergenerational update strategy based on consensus community inseting.Based on the non-dominant sorting strategy,the"non-dominanted particles" in the particle swarm are selected and the consensus communities are extracted,which guide the evolution direction of the particles and accelerate the convergence of the algorithm.At the same time,the implementation of the algorithm is simple and the computational complexity is low.Our method has good results even in large-scale networks.Finally,the proposed method is applied to protein-protein interaction networks,which is helpful to understand the structure,organization and function of cells from the perspective of biochemistry,signal transduction and genetic networks.The performance in two datasets of yeast shows higher recall and accuracy compared with the other six classical clustering algorithms on protein-protein interaction networks.Our method can effectively identify biological functional modules,which provides thoughts on the application of community discovery algorithm in high-throughput biological networks.