Automatic Question-Answer Generation from Educational Texts

As online educational platforms and blogs such as Coursera, Edublog, Korbit Technologies Inc. etc. are becoming increasingly popular, a great way for students to learn is to solve QA problems on educational texts. Students also learn by getting answers to their questions. Since manual QA generation of huge content of educational texts seems impracticable, an important research goal is to create natural question-answer generation systems from reading comprehension materials. Neural sequential models (e.g. LSTMs, Transformers etc.) are able to generate natural questions by paying attention to important part of texts and have improved state-of-the-art on many QA datasets, and also helped develop new datasets. In this research project, we will investigate how we can leverage current state-of-the-art NLP neural models for question-answering generation. We are equally interested in identifying lesser worked areas such as conversational QA (which helps develop intelligent agents that can drive QA style conversations and test user understanding of passage), exploiting active learning by querying the student for questions in-case of insufficient labelled data), crowdsourcing educational blogs for use in building QA models.

Intern: 
Devang Kulshreshtha
Faculty Supervisor: 
Siva Reddy
Province: 
Quebec
University: 
Partner University: 
Discipline: 
Program: