Order Sensibility of Spectral Clustering on Input Data Sets
谱聚类算法对输入数据顺序的敏感性*

Using tools from the matrix analysis,matrix representations of the spectral clustering algorithms was given.The corresponding matrixes getting from the different input order of the data set were similar to each other,so do the Laplacian matrixes.Then the matrixes Y whose columns were the top k eigenvectors of the Laplacian matrixes were similar to each other also.K-means algorithms using Y's row vectors as the clustering data set was one of the initialization-dependent algorithms.Then get the reason of why spectral clustering algorithms were sensitive to the input order of data set.