Examining the Feasibility of AI-Generated Questions in Educational Settings

Educators face ever-growing time constraints, leading to poor work-life balance and a negative impact on work quality. Through their language generation capabilities, large language models offer an interesting avenue to ease this academic workload, allowing both students and lecturers to generate educational content. In this work, we leverage the latest developments in automatic speech recognition, natural language generation, retrieval-augmented generation, and multimodal models to design the Augmented Lecture Integration Network (ALINet), a system capable of producing a diverse range of high-quality assessment questions from lecture content. We inform the design of our system through a series of automated experiments using public datasets and evaluate it with a user study conducted on students and educators. Our results indicate a generally positive perception of the system’s performance, particularly in generating natural and clear questions relevant to the taught content, demonstrating its potential as a valuable resource in educational settings. This project lays the foundation for future research in multimodal educational question generation and is available for reuse in our public repository.