E-QGen: Educational Lecture Abstract-based Question Generation System
Mao-Siang Chen, An-Zi Yen
TL;DR
The paper tackles the challenge of generating student-pertinent questions to aid instructors in lecture preparation. It introduces E-QGen, a system that maps lecture abstracts to educational transcripts and, through multitask learning with LoRA fine-tuning, produces three classes of questions: actual, probable, and potential. A large-scale dataset is built from MIT OCW and Stanford YouTube transcripts and comments, organized into golden, silver, and platinum pairs via transcript segmentation with text tiling and embedding- and classifier-based alignment, and augmented with GPT-4-generated platinum examples; the training optimizes Theta with a base model Phi_0 and increment DeltaPhi(Theta). Experimental results show that E-QGen, especially with the reference-question generator, achieves higher similarity and diversity to student questions than baseline models including GPT-4, and ablation confirms the value of platinum data. Overall, the work demonstrates a viable path to proactive, domain-specific question generation for educational contexts, with potential for expansion to broader disciplines.
Abstract
To optimize the preparation process for educators in academic lectures and associated question-and-answer sessions, this paper presents E-QGen, a lecture abstract-based question generation system. Given a lecture abstract, E-QGen generates potential student inquiries. The questions suggested by our system are expected to not only facilitate teachers in preparing answers in advance but also enable them to supply additional resources when necessary.