A Community Detection Algorithm In Complex Networks Based On Multi-objective Optimization

As the birth of the computer in twentieth Century,promoting the development of the Internet technology with lightning speed,Human society has entered the era of network information and imperceptibly been surrounded by a complex range of data.Countless networks are integrated into our lives with tangible and intangible way,from the telecommunications network,traffic network,aviation network,news network to social network,economic network,military networks,etc.These networks are some examples of complex networks,which have the common features of a wide variety of structures,complex structure,huge data,with a certain dynamic,a certain degree of self-organization and self-similarity in internal structure,and so on.Community structure is an important feature of complex network and it is an important way to understand the structure of a network.How to effectively detect the community structure of the network has a very important practical significance.In this paper,the community discovery problem is described as a multi-objective optimization problem(MOP).The community discovery algorithms based on multi-objective optimization in the presence exist some problems such as high computational complexity,lack of diversity and lack of local search.Aiming at these problems,in this paper,the main direction of improvement is multi-objective evolutionary algorithm,gene coding,genetic operation,optimization index and local search,etc.The algorithm is divided into two stages.First,the objective function of RC and MRA and genetic operations are determined and some strategies in the MOEA/D framework are improved,and it is optimized to obtain the Pareto optimal solution set for MRC and RA,corresponding multi-objective tradeoff and community structure.Secondly,in the model selection phase by modularity and NMI method,the best community division can been obtained.In the first phase,gene encoding based on the adjacency matrix is improved,so that each gene location has a label with two attributes,such as LNC which is the community of a node and LND which is the node fitness.In the process of decoding,LNC and LND are assigned meanwhile and the number of communities is automatically obtained.In this paper,we adopt two point crossover and gene mutation genetic operators based on node fitness,and set up h which is the threshold of node fitness.In the process of genetic operation,only the node fitness is less than the threshold,the corresponding gene location carrying out genetic operations,or not.This will reduce the randomness and the failure of the genetic operations and reduce the invalid individuals in the population,so that facilitating the evolution of a favorable population.In the IMOEA/D algorithm,the local search algorithm based on mutation and the selection strategy of genetic operations are introduced,which is helpful for the diversity of solutions.Experiments are carried out on the LFR benchmark and four real data sets,such as Karate Club?College Football?Bottlenose Dolphins and Kreb's Books.The experimental results show that the proposed algorithm can achieve better expected results.And it is better than other algorithms with strong stability and low complexity.