MCQBot, or Multiple Choice Questions Bot is a simple NLP project I realized in June 2018. The idea is pretty simple, given a question and a set of answers, make a bot that can gives a confidence score to each choice in order to reply to the question. Number of choices is not fixed.
Current method is pretty simple, n-grams occurrences-based, from google search content, but working quite efficiently. You can find the code on my github here
I have been running my bot over a small dataset of french questions and accuracy is about 85%.
In future work, accuracy could be increased using semantic analysis or smarter search of content using different source (such as Wikipedia for example), and/or different method (key words search rather than complete question search).
Finally, such “close field” question answering method could be expanded to “open field” questions by generating a range of potential choices.
For example, for questions such as Which european country ... ?, we could generate the list of choices (all european countries), and score each of them using this method.