StuQuestions

Overview

Asking questions forms an integral part of the learning process. However, online learning has reduced much of the interaction between teachers and students. This makes it harder for students to engage with and grasp the content while also making it challenging for teachers to gauge their students’ understanding. One viable solution to this is to create an Automatic Question Generation (AQG) system capable of generating questions from lecture transcripts. Such a system could be used to create a set of questions, from the lecture content, that can be used to access the students’ knowledge of the subject and to anticipate questions that students may have about the content.

Our research explores different approaches to AQG, a Data-Driven approach (Neural model) and a Rule and Template-Based approach.

Systems

Neural Question Generation

This research explores fine-tuning a language model (T5-base) with a small, diverse Information Retrieval (IR) dataset for question generation from lecture transcripts. It compares this model to one trained on a larger IR dataset (docT5query). Additionally, both T5-base and docT5query models are fine-tuned on the extensive education dataset LearningQ, along with five Reading Comprehension datasets, before fine-tuning on a small lecture transcript and questions data, in an attempt to improve the overall quality of generated questions, in terms of grammatical correctness, logical sense and relevance to the context. These fine-tuned models are then evaluated against the baseline.

Learn More

Template-Based Question Generation

This study focussed on developing two systems. A Rule-Based Semantic and a Template-Based Question Generation (QG) System. The main differentiating feature of these two systems is how their rules and templates are devised. The semantic system makes used of 8 manually created rules, while the template-based system combines semantic role labelling (SRL) with coded logic to automatically extract templates from sample input questions. The two systems are then compared in terms of question quality and dataset coverage. The template-based system is built with the goal to maintain the question quality produced by the manually created rules, while increasing the coverage compared to the semantic QG system. This aims to alleviate the tedious work assosiated with creating templates.