[seanp2k2]

I'd think you'd have better luck using bayes with some clever stemming / tokenization :)