Research map / Large language models / Language, safety and interpretability
How Teachers Can Use Large Language Models and Bloom’s Taxonomy to Create Educational Quizzes
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Generates reading-quiz questions from Wikipedia passages with GPT-3.5 using Bloom's-taxonomy-controlled few-shot prompts, then tests with 24 teachers whether quizzes written with the generated questions match handwritten quality and how teachers use them.
Problem
Educational question generation is rarely evaluated with real teachers or at the quiz level, so it is unclear whether LLM-generated questions are useful for teachers writing quizzes.
Approach
Prompt GPT-3.5 with five-shot examples that generate one question per Bloom's taxonomy level for each passage, generating all six at once to reduce duplicates, and compare teacher-written quizzes with and without these candidates in controlled quiz-writing experiments.
Key results
- Quizzes written with generated questions match handwritten quality; coverage increased significantly for controlled quizzes versus handwritten in both cohorts.
- In the ML cohort, usefulness ratings for simple quizzes were significantly higher than handwritten quizzes.
- Teachers strongly preferred controlled (Bloom's-based) generations and copied them more often than simple generations.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use automatically generated questions while writing quizzes | Teacher-written quizzes · BIO and ML passages | coverage | Helped |
| Use simple GPT-3.5-generated candidates while writing quizzes | Teacher-written quizzes · BIO and ML passages | quiz quality | No clear effect |
| Use Bloom's-taxonomy-controlled GPT-3.5-generated candidates while writing quizzes | Teacher-written quizzes · BIO and ML passages | quiz quality | Helped |
| Use generated candidates while writing quizzes | Teacher-written quizzes · BIO and ML passages | time to write quiz | No clear effect |
| Provide Bloom's-taxonomy-controlled generations instead of simple generations | Teacher quiz-writing process · BIO and ML cohorts | teacher preference | Helped |
| Provide Bloom's-taxonomy-controlled generations instead of simple generations | Teacher quiz-writing process · BIO and ML cohorts | directly copied generated questions | Helped |
| Use simple instead of controlled generated candidates | Teacher-written quizzes · ML cohort | usefulness | Hurt |
| Use generated candidates while writing quizzes | Teacher-written quizzes · BIO and ML passages | irrelevance, disfluency, and unanswerability | No clear effect |
Surprising findings
- Writing time was not consistently affected by using generated questions, and teachers wrote fewer questions from scratch when given controlled generations.
Limitations
- Quiz writing setting is contrived and may not reflect real teacher workflows.
- Only one LLM, two domains, English language, and a limited number of teachers.
- No student perspective is included.
Experimental setup
Datasets: Wikipedia passages (24 passages, biology and machine learning)
Metrics: coverage structure redundancy usefulness relevancy fluency answerability time taken question source (copied, edited, handwritten) teacher preference
Baselines: handwritten quizzes simple prompting strategy
Builds on: Bloom's taxonomy few-shot prompting prefix style prompting
Closest papers by meaning
- LLMs in the Classroom: Outcomes and Perceptions of Questions Written with the Aid of AI (AAAI 2025 · similarity 0.86)
- Explicitly Guided Difficulty-Controllable Visual Question Generation (AAAI 2025 · similarity 0.86)
- QGen Studio: An Adaptive Question-Answer Generation, Training and Evaluation Platform (AAAI 2025 · similarity 0.86)
- Exploring Iterative Enhancement for Improving Learnersourced Multiple-Choice Question Explanations with Large Language Models (AAAI 2025 · similarity 0.85)
- Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation (ICML 2024 · similarity 0.85)
- What Is a Good Question? Assessing Question Quality via Meta-Fact Checking (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.