A Complex Network Eigen Spectrum Analysis Based On Overlapping Community Detection

Community detection,as known as clustering in machine learning,has always been important research.With the increasing scale of data,the data structure becomes complicated,and the traditional clustering method has a strong dependence on the shape of data,which leads to the low universality.Spectral clustering(SC)combines the dimensionality reduction with the kmeans algorithm.Especially for high-dimensional data(such as text data),Spectral clustering has less computational complexity and higher clustering accuracy than k-means.Recently,spectral clustering has become one of the most popular clustering algorithms in machine learning and communication.With its advantages of easy implementation and wide application,the spectral method has been widely concerned by scholars.However,a large number of experimental results show that if there is noise in the data set,the spectral method often fails to provide satisfactory clustering results.At the same time,the data in the real world is inevitably noisy,so how to reduce the impact of noise on the spectral method is the key to the application of the spectral method in the real data set,and is also the focus of this paper.To address this problem,this paper starts from the Eigen spectrum,which is the core of the spectral method,to construct the model.The main work of this paper includes the following three aspects: 1)The framework;As we know the regularization method can effectively improve the anti-noise performance of the algorithm.In this paper,a regularized spectral clustering model based on entropy perturbation is proposed.which not only solves the problem of fixed-parameter but also improves the anti-noise performance of spectral clustering.2)Theoretical analysis;The algorithm in this paper is a matrix perturbation based method.Through matrix perturbation analysis and Davis-Kahan theory,this paper provides the theoretical threshold of the algorithm,which improves the universality and the theoretical background of the algorithm.3)Overlapping communities detection;At the end of this paper,we validated our algorithm by overlapping community detection,the experimental results on DBLP,Youtube,and other data sets prove the effectiveness of the proposed algorithm.