A QUESTION ANSWERING SYSTEM USING ENCODER-DECODER, SEQUENCE-TO-SEQUENCE, RECURRENT NEURAL NETWORKS

Publication Date

Spring 2018

Degree Type

Master's Project

Degree Name

Master of Science (MS)

Department

Computer Science

Abstract

Question answering is the study of writing computer programs that can answer natural language questions. It is one of the most challenging tasks in the field of natural language processing. The present state-of-art question answering systems use neural network models. In this project, we successfully built a question answering system using an encoder-decoder, sequence-to-sequence, recurrent neural network. In total, five different models were tried. The first model we implemented was a previously studied model called the Match Long Short Term Memory (LSTM) & Answer Pointer model. Our second, third, fourth and fifth models were designed by making changes to the first model to try to determine if all the components of this model were necessary. By comparing the results of these five different models, we found that two non-trivial weakenings of this model had minimal effect on the accuracy of the model in determining answers. On the other hand, when these weakening were both present the accuracy became substantially worse.

This document is currently not available here.

Share

COinS