다차원 메타데이터와 휴먼 인 더 루프 검증을 활용한 LLM 보조 반복적 토픽모델링: 메타버스 교육 사례 연구

초록

This study proposes an LLM-assisted iterative topic modeling framework that integrates multi-dimensional metadata structuring, multi-model consolidation, and human-in-the-loop validation. Research subjects, fields, and methods were extracted from academic documents, and LDA and BERTopic were applied in parallel to construct an initial topic set aligned with prior topic structures. The topic system was refined through selective cross-validation with heterogeneous LLMs, progressive sampling, and expert review. The finalized topic set was applied to the full corpus for distribution and trend analysis. The results show that metaverse education research has expanded around stable core themes rather than abrupt paradigm shifts. This study reconceptualizes topic modeling as an iterative process of semantic convergence and offers a structured empirical classification of metaverse education research.

키워드