E-QGen is a novel system that aims to assist educators in preparing for lectures and associated question-and-answer sessions. The system consists of three key components:
Educational Transcript Generator: This component automatically generates a complete lecture script based on the provided lecture abstract.
Student Question Generator: This component uses a multitask learning framework and LoRA fine-tuning to generate three types of questions: actual student questions, probable student questions, and potential student questions. The actual student questions closely align with what students typically ask, the probable student questions reflect topics students may care about, and the potential student questions estimate the inquiries students might have about course concepts.
Reference Question Generator: This component generates general conceptual questions to provide a more comprehensive set of course questions for the educators.
The authors constructed a dataset by collecting real student inquiries from publicly available lecture videos and transcripts uploaded by universities and research institutions. They leveraged language models to assist in the extraction and alignment of student questions with the corresponding lecture content.
Experimental results show that E-QGen outperforms other language models in generating questions that closely resemble those a student would ask, both in terms of similarity and diversity. The authors also conducted an ablation study to demonstrate the effectiveness of the multitask learning approach and the use of pseudo-training data generated by powerful language models.
The authors plan to extend the application of E-QGen to cover courses across various fields beyond computer science in the future.
In un'altra lingua
dal contenuto originale
arxiv.org
Approfondimenti chiave tratti da
by Mao-Siang Ch... alle arxiv.org 04-23-2024
https://arxiv.org/pdf/2404.13547.pdfDomande più approfondite