Chunk #15 — METHODS — SPECTRAL CLUSTERING
Text
graph Laplacian provide an approximate solution to the Ncut minimization problem. Smartpca [Patterson et al., 2006] and standard GEM [Luca et al., 2008], on the other hand, are biased toward embeddings that favor small, but tight, clusters in the data. For further exposition on this topic an excellent review is available [von Luxburg, 2007].