Structural Analysis of AI Liberal Arts Education Research in Korea: A Text-Mining Meta-Analysis of AI Liberal Arts Education Research in Korea

This study conducted a meta-analysis of 137 KCI-indexed journal articles on AI general education published in Korea between 2018 and 2025, using natural language processing-based text mining methods. To address the limitations of frequency-based analyses, including contextual loss and researcher subjectivity, the study applied TF-IDF keyword analysis, Sentence-BERT embedding with K-means clustering to identify thematic structures, and GPT-based cluster summarization to interpret the semantic characteristics of each cluster. The results revealed that research on AI general education in Korea is structured into six thematic clusters: instructional design using generative AI, AI literacy and foundational competencies, teacher AI competencies and professional development, effect studies examining learners’ experiences, emotions, and motivation, technology acceptance and attitude exploration, and discussions on AI ethics, responsibility, and sociocultural implications. Among these, studies on generative AI-based instructional design, writing, and critical thinking accounted for the largest proportion, whereas research focusing on ethics, responsibility, and social implications was relatively limited, indicating a practice-oriented research tendency. Analysis of inter-cluster relationships showed a relatively coherent cycle between instructional design and effect studies, while foundational competency education, teacher competency research, and ethics-related studies accumulated in a more fragmented manner. By employing a text mining-based meta-analytic approach, this study systematically maps the structural landscape and semantic network of AI general education research in Korea, providing foundational evidence for future curriculum design, teacher professional development systems, and ethics-oriented AI education policies. Nevertheless, the focus on article abstracts limits the extent to which the full research contexts could be captured, suggesting directions for future studies.