Automated Question Generation

Synthesis

Automated question generation leverages NLP and LLMs to create educational assessments at scale. Wei & Stamper (2025) introduced the generate-then-validate paradigm, reducing hallucination by 62% compared to direct generation and achieving 89% accuracy on STEM datasets.

Generate-Then-Validate Paradigm

1. Generation Phase: LLM produces candidate questions from source material 2. Validation Phase: Separate verification step filters invalid/low-quality items 3. Refinement Loop: Failed items trigger re-generation with corrective prompts

Advantages Over Direct Generation

Reduced Hallucination: Validation catches factually incorrect questions

Higher Relevance: 23% improvement on relevance metrics vs. baseline LLMs

Scalability: Enables rapid creation of formative assessments across domains

Educational Applications

Formative Assessment: Just-in-time questions for adaptive learning systems

STEM Education: Validated on mathematics and science problem generation

Differentiated Instruction: Generating multiple difficulty levels automatically

Connections

generate-then-validate-question-gen — Detailed implementation of the paradigm

nlp-education — NLP techniques for educational applications

automated-assessment — Reducing manual authoring burden

stem-education — Primary validation domain for question generation

llm-application-education — Generative AI for learning materials

learnmate2-llm-adaptive-learning — StudyMate could integrate question generation

References

Wei, Y., Stamper, J., et al. (2025). Generate-Then-Validate: A Novel Question Generation Approach. arXiv:2512.10110.

kt4eqg-personalized-question-generation — 7 of 8 papers in May 28 scan

slidesqaqa-pedagogical-question-generation — four-stage LLM pipeline with deck-level pedagogical reasoning

llm-handwritten-math-grading — Extends automated assessment research to handwritten math with vision-capable LLMs

taklif-ai-interest-based-personalized-assignments — LLM-generated personalized assignments

nsmq-riddles-science-math-benchmark — Riddle format as AI-generated educational content benchmark