The purpose of this project is to develop a Question Answering System with Reading Comprehension ability on Vietnamese, whose tools and resources are lacked, and applied to answering question related to rules and regulations of University of Information System.

This system adapts traditional Information Retrieval techniques (mostly based on Extended Boolean Model) and Deep Learning algorithms (BERT that achieves state-of-the-art performance on 11 different NLP tasks in English) and transfer learning on Vietnamese that posed attractive potential on Vietnamese Question Answering researches.

The Information Retrieval aprroaches are very common, but deep learning approaches are almost never used in Vietnamese QA System. In this project, a naive transfer learning technique is used, where we translate the SQuAD dataset from English to Vietnamese and remove bad translation (link) that poses an additional 10% boost in F1 accuracy, resulted in an F1 accuracy of 66% in the original task (QA on Wikipedia) and 56% in the UIT regulations task.

More information about this project is stored in ./Report/Summary.pdf or ./Report/Thesis.pdf

• QASystem and Ultilities contain source codes, base model as well as fine-tuning models and dataset used in this project. Guide on how to setup and re-produce the result is also provided.
• Report contains documents about this thesis as well as slides and related files.
• Dataset contains the dataset that is used in this project.

• By Nguyễn Việt Nam - 14520560
• Advisors: Dr. Ngô Đức Thành & M Sc Nguyễn Vinh Tiệp
• Advanced Education Program 2014 - VNU-UIT

If any problem occurs, please contact me via my email address [email protected] or [email protected]