Friends And Relatives Call Circle Mining Based On Spectral Clustering

With the rapid development of the Internet,Data on the network grows exponentially. Theflood of information resource contains a large number of useful information for us,butunfortunately,it is difficult to find the information we really need quickly and efficiently. Datamining skill can find valuable information for us from large amounts of data,and clusteranalysis is a classic problem in the area of data mining.Cluster analysis is a very active research hotpot in the field of data mining and machinelearning， it is a effective means for people to understand and explore the intrinsic linkbetween things,and spectral clustering algorithm is an important branch of cluster analysis.Spectral clustering algorithm is based on the theory of spectral graph theory which is theory todivide the data set.Spectral clustering algorithm as a novel clustering analysisalgorithm,compared with traditional clustering methods,spectral clustering has obviousadvantages. It is not only easy to understand, easy to implement, not easy to fall into localoptimal solution,can identify the distribution of non-convex,and it can cluster on any samplespace and converge to the global optimal solution,so it is ideally suited for many practicalapplication problems.The traditional spectral clustering algorithm has multiple versions,Thebiggest difference between them is the choice of different Laplacian Matrix.The traditionalspectral clustering algorithm first obtain the Laplacian matrix by data set,Then come to thefirst k Laplacian matrix eigenvalues and eigenvectors,construct feature vector space,then usethe K-means clustering algorithm, or other classic clustering algorithm on the feature vectorspace.Firstly, this paper analyzes the family relations and determines the relevant attributes andpreprocess the data，and then analyzes the theory and methods of spectral clustering，thereason of using spectral clustering and the application advantages of it,and point out someprevalent problems of current spectral clustering.Through analyzing the problem,proposing aimproved solution,then use the improved algorithm to mine the relatives and friends callcircle and validate the improved algorithm, and analysis the value of the results finally.