I had a lot of rhetorical questions too. Also, Kevin's solution is not 100% waterproof - just looked through some "negative reviews" for ServInt and they weren't exactly negative. Maybe you could build a bayesian filter of some kind? (I'm not sure how Kevin filtered his reviews exactly?)
Maybe he could add a third button, "Not a question", classify each question into "answerable question" and "not a question" categories (according to which button the user clicks) and see which category new questions are classified into using a naive bayes classifier? I think it could work quite well
Kevin here. No, it's not perfect and never will be. I definitely use a bayesian classifier in our process among other things.
That said, if you see mistakes, please flag them! My system does get smarter with more data, so when people flag mistakes for us to correct, it makes a difference both immediately and in the long run.
If you have any other thoughts/critiques/ideas please feel free to let me know :)
