Research On Community Detection Algorithm For Complex Networks Based On Intelligent Algorithm

Complex network is an important expression form of complex system,which is helpful to study the topological structure of complex system.With the application of complex networks in social,biological,transportation,and information fields,community structures,as one of the important characteristics of complex networks,can reveal the functions and properties of nodes in the network and are widely used in life.For example,community structure technology is applied to drug trafficking network to screen out important drug traffickers,search engine response speed issues to aggregate the relevance of web pages,protein network to predict protein functions,etc.In recent years,community detection algorithms have been proposed one after another,but there is still low accuracy of community division,and with the continuous expansion of network scale,the traditional community discovery methods are not competent.To solve these problems,this paper proposes a community detection algorithm based on improved bee evolutionary genetic algorithm and based on depth encoder algorithm according to different scale of complex networks.(1)Aiming at the problem of low accuracy of small-scale networks,this paper proposed a community detection algorithm based on an improved bee evolutionary genetic algorithm.Based on the advantages of genetic algorithm optimization,the bee evolutionary genetic algorithm was used for the first time to explore the community structure.Firstly,the algorithm took modularity as the fitness function,combined the improved character encoding method with the corresponding genetic operators,automatically acquired the optimal community number and the community detection solution without the prior knowledge.Then,by utilizing the local information of the network topology structure in initialization population,crossover operation and mutation operation,the search space was compressed,the optimization ability and convergence speed was improved and introducing the number of random population inversely proportional to the number of iterations to improve the exploration ability,robustness and accuracy of the algorithm.Finally,the proposed IBEGA was tested on real networks.The results show that compared with other classical algorithms for community discovery and similar intelligent algorithms,the algorithm has the advantages of high accuracy,which shows that the algorithm is feasible and effective.(2)Aiming at the problem of high time consuming and low accuracy in large-scale network,this paper proposed a deep auto-encoder and Eforest encoder algorithm and effect diffusion similarity index.This algorithm combined a multi-layer auto-encoder with a EForest to form a two-level cascade model,and transformed the similarity matrix into low dimension and higher order feature matrices through dimensionality reduction and characterization learning,and used K-means to obtain community detection results.The cascade structure greatly reduced the time complexity of the model while maintaining the same depth of the algorithm.The simulation result show that the proposed algorithm was more accurate than the K-means,Spectral,DA-EML and Co DDA on synthetic data sets and real data sets.In the experiment of algorithm performance,the rationality and validity of the cascade structure of the algorithm,the depth of the auto-encoder and the effect diffusion index are verified.