@article{ART003370763},
author={Ki-Sung Song},
title={A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={8},
pages={171-179}
TY - JOUR
AU - Ki-Sung Song
TI - A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 8
PB - The Korean Society Of Computer And Information
SP - 171
EP - 179
SN - 1598-849X
AB - Spatial data are continuously used in policy decision-making, but they are difficult for non-experts to use because spatial relationships and location-based operations must be considered together. Natural language spatial query technologies can serve as a technical alternative that enables users to directly specify required analytical conditions beyond predefined dashboards or prewritten query statements. This study compares the effects of metadata configurations on NL2GeoSQL execution accuracy using Seoul’s crime-prevention decision support data. To this end, a PostGIS-based spatial database and an evaluation set of 30 Korean natural-language spatial queries were constructed, and SQL statements generated by two code-focused large language models under five metadata conditions were executed and evaluated.
The results show that the SQL-to-text example condition achieved the highest execution accuracy and also performed relatively well in schema linking and spatial function use. This study demonstrates that metadata configuration affects GeoSQL generation performance and that SQL-to-text examples can be used as execution-oriented knowledge for GeoSQL generation.
KW - NL2GeoSQL;Metadata;Spatial Database;Large Language Model(LLM);Execution Accuracy
DO -
UR -
ER -
Ki-Sung Song. (2026). A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database. Journal of The Korea Society of Computer and Information, 31(8), 171-179.
Ki-Sung Song. 2026, "A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database", Journal of The Korea Society of Computer and Information, vol.31, no.8 pp.171-179.
Ki-Sung Song "A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database" Journal of The Korea Society of Computer and Information 31.8 pp.171-179 (2026) : 171.
Ki-Sung Song. A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database. 2026; 31(8), 171-179.
Ki-Sung Song. "A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 171-179.
Ki-Sung Song. A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database. Journal of The Korea Society of Computer and Information, 31(8), 171-179.
Ki-Sung Song. A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database. Journal of The Korea Society of Computer and Information. 2026; 31(8) 171-179.
Ki-Sung Song. A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database. 2026; 31(8), 171-179.
Ki-Sung Song. "A Comparison of NL2GeoSQL Execution Accuracy Across Metadata Configurations: A Case Study of Crime Vulnerability Grid Database" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 171-179.
