Development of an LLM-based TOPIK Item Generation Model and Its Validity Verification

The Hallyu (Korean Wave) content industry has become a core axis of the global cultural market, spurring increased interest in Hangeul and the Test of Proficiency in Korean (TOPIK). However, following the 2020 Korean Standard Education Curriculum, the supply of official and private practice items remains limited, hindering the ability to secure sufficient mock exam materials. To solve this, this study successfully developed an automatic item generation (AIG) model and system that adheres to specialized TOPIK standards, including item-specific learning objectives. The system’s performance was rigorously verified through evaluations conducted by Korean education experts, confirming the alignment of generated items with their respective learning objectives. Critically, the model addresses grammatical and sentence structure errors resulting from the data sparsity of Hangeul in commercially available foreign LLMs. By automating the creation of all item components (passages, options, commentary, and attractive distractors) and ensuring they meet TOPIK’s inherent level-specific objectives, the AIG system significantly enhances the overall quality and completeness of mock exams.