Research And Application On Community Detection Algorithms In Complex Networks

Community structure,which consists of nodes sharing the same attributes in the network,is one of the most important feature of complex networks and plays a crucial role in the real system of representation.Investigating community structure in the network has profound significance for helping to analyze the properties of the network function.At present,many community detection algorithms have been widely used in complex network community to discover problems.How to reduce the complexity of algorithm and improve the accuracy of community segmentation results is always a developing trend and research direction of community detection algorithms on the grounds of a directed-weighted network.Firstly,a quantum-behaved discrete particle swarm optimization for complex network clustering is proposed in non-overlapping community detection algorithm(NQD-PSO).The core node and neighborly common nodes are constructed as motif,which is the initial value of the quantum particle swarm optimization algorithm.At the same time,constructing the motif weighted community clustering function as the adaptive function of the algorithm,and the compression factor is adopted to adjust the global and local search model,which makes the algorithm globally converge by combining with quantum particle swarm optimization.Compared with other algorithms,NQD-PSO algorithm uses motif orderly table coding method,and experimental results on both synthetic and real datasets show that.the NQD-PSO algorithm can mine more high-quality community structures.Secondly,a local extended genetic algorithm optimization overlapping community detection is proposed in this study(LEGAOCD).It makes a few core nodes constructed as die body,yet regards the main idea of local extended overlapping community detection as reference;meanwhile,constructing the core node weighted community clustering function as the adaptive function of the algorithm,while it can use the triangular model to judge the problem of community stability measurement for quantifying the stability of community.Then,the improved strategy of genetic algorithm is used to allocate the communities where they belong and the core node ordered table coding method is used to achieve rough community structure.Finally,the high-quality overlapping community structure is obtained by two discriminant objective functions.After that,the LEGAOCD is compared with classical CPM and COPRA algorithms on the data sets,the results show that LEGAOCD possesses better in the aspects of detecting overlapping community structure and overlapping nodes.Finally,those algorithms are extended to the financial network model,which belongs to the heterogeneous directed weighted network model.A directed-weighted financial network model is built to take such stocks as nodes that had been traded uninterruptedly in Shanghai A-share stocks market and taking the correlation of stock the closing prices fluctuation as edges and weights.The network model is preprocessed by random matrix denoising method,it is found that denoising network still remains crucial information on topology of the original networks by analyzing the original network and topology of denoising network.Besides,according to the effective threshold,same directed-weighted financial network model are built with different thresholds.Those algorithms are used to mine community respectively for financial network model,compared with several state-of-the-art algorithms,the results show that the division of community structure of the NQD-PSO algorithm and the LEGAOCD algorithm are clearer and their divided structure are more compat.