Spectral Clustering Based Overlapping Community Detection In Complex Networks

In many real-world systems,die relationship between objects and objects can be modeled as complex networks for analysis.The modular structure can usually explain the topology and functional modules of complex network systems,and is an important attribute of complex networks.The complex network community detection aims to explore the modular structure in such a system with complex network structure.Studying this modular structure helps to better understand and explore the hidden functions of the network system.For many years,researchers in various fields have proposed a number of community mining algorithms,and conducted in-depth research on complex network community detection in different subject areas.With the concept of overlapping community structures,these algorithms are in the field of overlapping community discovery.Woth further research.Therefore,the traditional community detection algorithm based on spectral clustering can not mine the problem of overlapping community structure.This thesis proposes a spectral clustering overlapping community detection algorithm based on edge partitioning.In addition,the community structure of complex networks is not the same.In order to make the algorithm have better robust performance in different structures,this thesis further proposes the ensemble overlapping community detection algorithm based on spectral clustering.Both algorithms are based on the idea of the spectral clustering algorithm to study the overlapping community detection algorithm.The main research works of this thesis are as follows:(1)A spectral clustering overlapping community detection algorithm based on edge partitioning is proposed here.At present the community detection algorithm based on spectral clustering can process complex network structures well when detecting non-overlapping community structures,and obtain more accurate classification results,but it can not solve the problem of overlapping community detection well.The main reason is the spectral clustering method derives the new features of the network nodes by matrix spectrum analysis theory,and uses the new data features to cut the original network into k subgraphs that are not connected to each other,so the overlapping community structure cannot be detected.Therefore,this thesis proposes a spectral clustering overlapping community detection algorithm based on edge partitioning.The main idea is to combine the idea of spectral clustering algorithm with the idea of edge partitioning,and to construct the inner community of the network by constructing the similarity matrix between edges and edges.The structure is to mine the non-over-lapping nodes and the candidate overlapping nodes in each community,and then gradually analyze the overlapping nodes by analyzing the neighbor node information of each community candidate overlapping node to obtain the final overlapping community partitioning result.The experimental comparison on the artificial LFR network proves that the method can effectively mine the complex overlapping community structure in the network and can effectively solve the problem of excessive overlap of edge division.(2)A complex network clustering ensemble overlapping community detection algorithm based on spectral clustering is proposed here.At present,the complex network community detection algorithm based on spectral clustering is simple to implement and is not easy to fall into local optimum.However,the computational complexity of this method is relatively high,and it depends on the selection of community scale parameters.The main reason is that with the increase of network scale.The feature vector cannot achieve sufficient accuracy due to the high computational complexity.In order to solve this problem and improve the robustness of the algorithm,this thesis proposes a complex network clustering integrated overlapping community detection algorithm based on spectral clustering.The main idea of this method is to extract a plurality of representative points from each community as much as possible by designing a new sampling method,and perform spectral clustering division of sampling points according to the perturbation theory of the matrix,so as to mine nodes in the network.The community information between the two is then weighted and integrated according to the topology of the network and the quality of multiple divisions to obtain a weight network,thereby strengthening and mining the network community structure and obtaining the final overlapping community division.By comparing several existing overlapping community detection algorithms on LFR generation networks and real networks with different complex structures,the algorithm can be stable and good in multiple networks with different complex structures.Robust overlapping community partitioning results.And through further simulation experiments,the algorithm can sample the representative nodes from all communities well,and the integrated netxwork has a good community structure strengthening effect.