Comparative Question Answering System based on Natural Language Processing and Machine Learning

Abstract

It is very tedious for anyone to go through the whole document to get answers for their queries since there is a need of Question Answering system to make life easier. In this research, we used machine learning architectures in Question Answering field, based on the Stanford Question Answering dataset (SQuAD). In our work, build two models in which first model used unsupervised learning algorithm GloVe to get vector representation of word and trained using bidirectional-LSTM in which the training accuracy is 64.93% and testing accuracy is 60.33% scored with the model. In Second model used InferSent which is a sentence embedding method to get vector representation of data. This vector representation data is used to train model. The machine learning algorithms used are XG Boost and Multinomial Logistic Regression in which scored 70.02 percent in training and 66.03 percent in testing. The aim of this research is to build the best accuracy model using Glove and InferSent to use vector of various dimensionality to represent it numerically so that machine can interpret it.