거대 언어 모델을 활용한 일차방정식 문항의 오개념 기반 오답지 자동 생성 = Automatic generation of misconception–based distractors for linear equation items using large language models

다국어 초록 (Multilingual Abstract)

With the digital transformation in education, research on Automatic Item Generation (AIG) for building extensive item banks is becoming increasingly significant. In particular, generating effective distractors for diagnosing students’ learning deficiencies is crucial. This study explored methods for automatically generating misconception based distractors for linear equations in middle school mathematics using a Large Language Model (LLM).
Employing GPT–4o, a total of 321 distractors were generated based on five error types, including logical errors, and five item types, such as simple linear equations. Results indicated approximately 60% success rates for three error types, whereas other types showed lower or no success. By item type, math word problems had notably low success (10.4%), while others averaged around 30%. Some generated distractors exhibited limitations, such as excessive algebraic manipulations or unrealistic solutions. Future studies should enhance prompt design to better reflect students’ cognitive processes and discuss practical applications of generated distractors.