Build a question answering system that can compete against humans in a “quiz bowl”-like game. Create tf-idf vector representations of questions and answer documents (wiki pages). Create a KNN classifier with vector representation of questions as inputs to give an answer. “Automatically Buzz” for answers above a certain threshold value even before the whole question end. For every answer, we get a confidence measure [which is the cosine similarity measure of the vector representations] of the Question and the Answer document. If the answer is over a threshold for any sentence in the question, buzz and make it the answer. If no answer is above the threshold, buzz the answer predicted from the last sentence.

Share this project:

Updates