Trolling is the art of feigning one or more of the following: empathy, ignorance, sincerity. You're really only fending off non-native speakers and spammers who

Trolling is the art of feigning one or more of the following: empathy, ignorance, sincerity. You're really only fending off non-native speakers and spammers who give up very quickly (one could guess and ultimately get many correct answers). Not to mention that there will be a finite and probably short list of question/answer combos. Most of the variation will be in the questions I imagine and it will be in the form of noise (synonyms and different sentence structures). They make it very easy to scrape too. The question text is plain text (no need to OCR that) and you can run that through your favorite NLP package to extract useful structure to work from. The answers are each displayed as individual images on the page, making it trivial to know how many possible answers there are. That also makes it easier to run your favorite OCR package on the individually sliced images. Always glad to see people cracking away at this problem though. It's great fodder for machine learning enthusiasts.