Abstract:
Question-Answering systems are becoming increasingly popular, especially after the advent of Machine Learning. Such systems can be used in a wide variety of applications ranging from Open domain question answering or closed domain question answering which involves searching for answering within a fixed domain. This thesis explores the idea of using sentence matching as a technique to answer question based on a corpus of text provided to us, (similar to that of machine comprehension). In this thesis, we mainly focus on facts about Dr. A.P.J. Abdul Kalam as the corpus given to the system as input. Based on the Questions asked by the user, the system searches for correct facts that contain the relevant information. This work first explores corpus statistics as measure of sentence matching, followed by two different representation of word vectors to represent the sentences in the vector space,viz word2vec and fastText. Next, we explore Siamese Deep Learning Network to perform sentence matching. Finally we compare the results with sentence vectors captured using sentence2vec.
