Automatic question generation (part B)

• Main Author: Ong, Kenneth Rong Qing
• Other Authors: Andy Khong W H
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering::Computer hardware, software and systems; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/158051
• Institution: Nanyang Technological University

Question Generation System aims to help education system by generating deep questions that allows the students to think critically before a correct answer may be obtained. As more online courses are being conducted, education system is heavily reliant on ensuring that quality questions can be generated to help in students’ learning, as well as for assessment purposes. This project focuses on identifying the baseline paper for deep question generation, followed by the implementation to benchmark with the reported results and lastly to propose recommendations that can potentially improve the baseline results. The dataset that will be used for this implementation is known as the HotpotQA, a dataset that was created for the question answering domain that can be interchangeably used for question generation system after performing modifications. For this project implementation, the core lies in the construction and use of the dependency parsing semantic graph to represent the information of the dataset used. After the construction of the semantic graph, attention mechanism will be deployed to distinguish the important nodes in the graph that are critical for deep question generation. The semantic graph will be subsequently used to train a classifier for node classification task. At the training stage, the classifier will be used for joint training with the question generator for deep question generation task. The question generator that has been trained will therefore be able to perform deep question level prediction. Finally, evaluation of the generator model is conducted to measure the quality of the predicted questions using Bilingual Evaluation Understudy (BLEU) 1, 2, 3, 4, Metric for Evaluation of Translation with Explicit Ordering (METEOR) and ROUGE L.