Automatic Generation of Personalized Feedback in Intelligent Tutoring Systems

As Intelligent tutoring systems such as Coursera, Korbit become popular, a great way for students to learn is by receiving feedback for incorrect answers. Since manual feedback generation on large educational texts seems impracticable, an important goal is to create hint generation systems from student-teacher conversations. These systems can be deployed as chatbots giving hints on wrong answers, and annotated datasets for NLP research in dialogue-based question answering.
Initial solutions focused on designing rule-based approaches, typically based on deep linguistic knowledge. Although producing useful hints, these methods rely heavily on manual labour.
In this project, we leverage neural NLP models (like BERT, BART etc.) for automatic generation of pedagogical interventions. We are equally interested in less explored domains – active learning by asking the student to select correct hints, conversational QA (to develop intelligent agents driving QA style conversations). Another important direction is to use hint generation for improving tasks like question-answering, question generation etc.

Faculty Supervisor:

Siva Reddy

Student:

Partner:

Korbit Technologies

Discipline:

Computer science

Sector:

Information and cultural industries

University:

McGill University

Program: