Weighting will be a challenge, but is a good idea, thanks. I'll take a look at some NLP libraries and see if this can be improved.