The reaserch on parameters of spectral clustering based on SVD

The spectral clustering algorithm, which cluster by using the eigenvalues and eigenvectors of Laplacian matrix, may not be obtained the desired clustering results in some cases. It is possible to remedy this deficiency by using the singular value decomposition (SVD) in the spectral clustering algorithm. It is presented in this article the algorithm of spectral clustering based on SVD which use singular value decomposition (SVD) instead of Laplacian eigenmaps. The choice of singular vectors, the estimate of the Gaussian kernel parameter, and the determine of the clusters number significantly influence the clustering results. The experiments show that these spectral clustering parameters such as the number of clusters, etc. can be calculated by use of the dimensionality reduction factor and the expression of singular value.