This study explores how integrating Knowledge Organization Systems (KOSs) with generative artificial intelligence (generative AI) can enhance retrieval and discovery processes. KOSs offer structured vocabularies to support knowledge organization, whereas generative AI provides advanced language processing and recommendation capabilities. A major challenge is hallucination, where AI generates responses that appear plausible but are factually incorrect. To address this issue, the study examines how the quantity and quality of metadata influence hallucination mitigation and recommendation accuracy. The study evaluates four levels of metadata (0–3) using an iterative approach that incorporates expert feedback for validation. The results indicate that higher levels of metadata improve the precision of recommendations and significantly reduce hallucination rates. In addition, refinement was applied to Level 0, where hallucinations were most frequent, and Level 2, where metadata usage was suboptimal, through prompt engineering and iterative feedback. The findings confirm that structured prompts and user feedback can effectively enhance the reliability of AI-generated recommendations. Moreover, the study emphasizes the essential role of expert involvement in curating high-quality metadata and ensuring the credibility of AI-driven knowledge retrieval systems. Future research should investigate multilingual KOS recommendations, examine the scalability of AI-integrated KOS models, and develop methods to ensure consistency and reproducibility in AI-generated recommendations.
Ziyoung Park (Tue,) studied this question.