Blog Comment Similarity Detector (Free Code For Disqus)

I used essentially this method as well for personas (personas.media.mit.edu) as so many web results were the same... except I kept track of duplicates to then employ further metrics to select 'the best one.'

Works well.

Although the tokenization is very much a poor mans. I have since open sourced my tokenization framework (for python) which handles real world text: http://bitbucket.org/azinman/defuse/src/tip/smgtk/tokup/

Aaron