Why Do We Need SVD Before Clustering?

[aha123]

Text Cluster node in Text Miner peforms SVD before clustering. Can anyone tell me the advantage of SVD here?

[JohnJPS]

I think this oft-cited paper (http://www.cc.gatech.edu/~vempala/papers/dfkvv.pdf) describes it as well as it can be explained. Basically, they talk about how clustering the SVD solves an approximate clustering solution for the actual dataset, with much better performance.  So it's probably that performance boost that is the primary explanation.

[aha123]

This paper is excellent. Do you have any paper about how much information is lost after doing SVD? PCA can tell you how much variance is kept in first x numbers of PCs. Wonder if SVD has such measurement. Also why not use PCA for text clustering purpose? I read online many articles and still can't get any clear answer.

[JohnJPS]

I dug a little deeper and this discussion really does a great job: starting with PCA and moving onto SVD: https://www.cs.princeton.edu/picasso/mats/PCA-Tutorial-Intuition_jp.pdf

A more brief Q&A that is quite nice is here: https://www.quora.com/What-is-an-intuitive-explanation-of-the-relation-between-PCA-and-SVD

Hope that helps!