heh, I recently built a naive bayes analyzer that decides whether a text is from reddit or from ycnews. The assumption was that Reddit-like comments/stories were fluff and not worth reading.