| Title |
Effects of Prompt Design on LLM-Based Land Use Planning - Experimental Comparisons of Component Ordering and Prompting Techniques |
| Authors |
남영현(Nam, Young-Hyun) ; 권영상(Kwon, Young-Sang) |
| DOI |
https://doi.org/10.38195/judik.2026.08.27.5.7 |
| Keywords |
대규모 언어모델; 인공지능; 토지이용계획; 프롬프트 엔지니어링; 베이지안 분석 Large Language Model; Artificial Intelligence; Land Use Planning; Prompt Engineering; Bayesian Analysis |
| Abstract |
This study aims to verify the impact of prompts on the land use ratio accuracy of generated output in the LLM-based land use planning design process. Using Gemini 3 Thinking Mode with Yeouido, Seoul as the study site, generated outputs were compared across the arrangement order of prompt components and prompt techniques. Each condition was repeated 30 times, and the accuracy of land use proportions in AI-generated plans was evaluated using RMSE, with frequentist and Bayesian analysis. Neither factor produced significant differences, and Bayesian analysis provided evidence in favor of the null hypothesis of no difference in both experiments. The significance of this study lies in providing evidence that the surface features of prompts do not significantly affect land use ratio accuracy. |