Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Context Selection and Rewriting for Video-based Educational Question Generation

Mengxia Yu, Bang Nguyen, Olivia Zino, Meng Jiang

TL;DR

This work tackles video-based educational question generation (EQG) by addressing the challenge of generating questions from long, noisy classroom transcripts and aligned slide content. It introduces COSER, a two-stage framework that first dynamically selects answer- and timestamp-relevant context from both transcripts and video keyframes, then rewrites this context into concise, answer-containing knowledge statements, and finally integrates multimodal information for question generation. To study real-world classroom settings, the authors present AIRC, a dataset consisting of live lecture recordings, transcripts, slide visuals, and educator-created timestamp-based MCQs across two university courses. They validate COSER using an NLI-based, reference-like evaluation metric and show that selective, rewritten, and multimodal contexts significantly improve question relevancy and answerability, with transcripts generally providing more useful context than keyframes and multi-modal integration yielding the best results. This work advances practical video-based EQG by enabling targeted, timestamp-aware context construction and by providing a realistic benchmark and code release for future research.

Abstract

Educational question generation (EQG) is a crucial component of intelligent educational systems, significantly aiding self-assessment, active learning, and personalized education. While EQG systems have emerged, existing datasets typically rely on predefined, carefully edited texts, failing to represent real-world classroom content, including lecture speech with a set of complementary slides. To bridge this gap, we collect a dataset of educational questions based on lectures from real-world classrooms. On this realistic dataset, we find that current methods for EQG struggle with accurately generating questions from educational videos, particularly in aligning with specific timestamps and target answers. Common challenges include selecting informative contexts from extensive transcripts and ensuring generated questions meaningfully incorporate the target answer. To address the challenges, we introduce a novel framework utilizing large language models for dynamically selecting and rewriting contexts based on target timestamps and answers. First, our framework selects contexts from both lecture transcripts and video keyframes based on answer relevance and temporal proximity. Then, we integrate the contexts selected from both modalities and rewrite them into answer-containing knowledge statements, to enhance the logical connection between the contexts and the desired answer. This approach significantly improves the quality and relevance of the generated questions. Our dataset and code are released in https://github.com/mengxiayu/COSER.

Context Selection and Rewriting for Video-based Educational Question Generation

TL;DR

This work tackles video-based educational question generation (EQG) by addressing the challenge of generating questions from long, noisy classroom transcripts and aligned slide content. It introduces COSER, a two-stage framework that first dynamically selects answer- and timestamp-relevant context from both transcripts and video keyframes, then rewrites this context into concise, answer-containing knowledge statements, and finally integrates multimodal information for question generation. To study real-world classroom settings, the authors present AIRC, a dataset consisting of live lecture recordings, transcripts, slide visuals, and educator-created timestamp-based MCQs across two university courses. They validate COSER using an NLI-based, reference-like evaluation metric and show that selective, rewritten, and multimodal contexts significantly improve question relevancy and answerability, with transcripts generally providing more useful context than keyframes and multi-modal integration yielding the best results. This work advances practical video-based EQG by enabling targeted, timestamp-aware context construction and by providing a realistic benchmark and code release for future research.

Abstract

Educational question generation (EQG) is a crucial component of intelligent educational systems, significantly aiding self-assessment, active learning, and personalized education. While EQG systems have emerged, existing datasets typically rely on predefined, carefully edited texts, failing to represent real-world classroom content, including lecture speech with a set of complementary slides. To bridge this gap, we collect a dataset of educational questions based on lectures from real-world classrooms. On this realistic dataset, we find that current methods for EQG struggle with accurately generating questions from educational videos, particularly in aligning with specific timestamps and target answers. Common challenges include selecting informative contexts from extensive transcripts and ensuring generated questions meaningfully incorporate the target answer. To address the challenges, we introduce a novel framework utilizing large language models for dynamically selecting and rewriting contexts based on target timestamps and answers. First, our framework selects contexts from both lecture transcripts and video keyframes based on answer relevance and temporal proximity. Then, we integrate the contexts selected from both modalities and rewrite them into answer-containing knowledge statements, to enhance the logical connection between the contexts and the desired answer. This approach significantly improves the quality and relevance of the generated questions. Our dataset and code are released in https://github.com/mengxiayu/COSER.
Paper Structure (39 sections, 1 equation, 2 figures, 7 tables, 1 algorithm)

This paper contains 39 sections, 1 equation, 2 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Educational Question Generation with LLM
  4. Context Modeling for Question Generation
  5. Dataset: AIRC
  6. Overview
  7. Data Collection
  8. Data Postprocessing
  9. Preliminary
  10. Answer and Timestamp-Aware Educational Question Generation
  11. Context Construction for ATEQG
  12. Method
  13. Context Selection
  14. Context Rewriting
  15. Multi-Modal Integration
  16. ...and 24 more sections

Figures (2)

  • Figure 1: Our proposed framework COSER. Using all lecture content (left) as context results in generated questions that are too general and fail to incorporate keywords. COSER (right), which (1) dynamically selects relevant contexts from both transcripts and keyframes, and (2) integrates and rewrites them into answer-containing knowledge points, yields more specific and relevant question.
  • Figure 2: Increasing context window does not always improve question generation.