Analysis of Research Trends in ‘AI Architecture’ Before and After the Diffusion of LLMs Using Text Mining

This study examines how the thematic structure and conceptual focus of “AI architecture” research in Korea have evolved before and after the diffusion of large language models (LLMs). Abstracts of papers published in the DBpia database between 2019 and 2025 were analyzed using text-mining techniques, including word frequency, TF IDF, network centrality, and CONCOR analysis. The results show that pre-LLM studies mainly addressed design, space, energy, and construction performance, emphasizing technological functionality and optimization. After LLM diffusion, keywords such as generative, image, intelligence, model, and BIM emerged, reflecting a transition toward data-driven, automated, and creative design processes. Network and clustering analyses further revealed that AI architecture research shifted from performance-oriented technical applications to open, multidisciplinary structures centered on AI-based creation and intelligent modeling. These findings demonstrate that the diffusion of LLMs marked a paradigm shift in AI architecture from tool-based application to conceptual reconstruction of the design act providing a foundation for future research on AI human collaborative and generative design.