Automatic question generation from freeform text

Abstract

Automatic question-answer pair generation has many potential applications in the areas of FAQ preparation, question-answer dataset creation, and education. This thesis is about how to automatically generate Short Question-Answer pairs, Multiple Choice Question-Answer pairs, Boolean Question-Answer pairs, and Question Answering from input text without knowing the answers in advance. In this research, we applied Natural Language Processing and deep learning technologies: the answers were extracted from the text by keywords extraction skills; Using the extracted answers and input text to generate the question is completed by the Text-to-Text Transformer model; Finding options for the Multiple Choice Question-Answer pairs is completed by sense to vector models. A web interface has been developed to demonstrate the results of these different kinds of Question-Answer pairs development. Keywords: Question-Answer pair generation, Text-to-Text Transformer model, Natural Language Processing, web- site development