Geographic Inference with Large Language Models: Identification of POIs and Urban Features from Addresses and Coordinates

• Gabriel A. C. Silva Universidade Federal de Campina Grande (UFCG)
• Salatiel D. Silva Universidade Federal de Campina Grande (UFCG)
• Claudio E. C. Campelo Universidade Federal de Campina Grande (UFCG)

DOI: https://doi.org/10.5753/sbbd.2026.249658

Resumo

This paper evaluates the ability of Large Language Models (LLMs) to infer geographic features and identify points of interest (POIs) from addresses and coordinates. Nine cities of different urban sizes, distributed across three continents, were analyzed, with five addresses per city in scenarios such as urban center, residential area, green area, proximity to water and mixed area. The predictions were validated with external geospatial databases. The results indicate greater stability in feature inference than in POI identification, whose performance varies according to country and urban size; Gemini 3.1 Flash Lite leads in large cities, GPT-4.1 leads in medium-sized cities, and Gemini 3 Flash Preview leads in small cities.

Palavras-chave: Geographic Inference, Large Language Models, Evaluation, Urban Features, Point of Interest

Referências

Bhandari, P., Anastasopoulos, A., and Pfoser, D. (2023). Are large language models geospatially knowledgeable? In Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems, SIGSPATIAL ’23, New York, NY, USA. Association for Computing Machinery.

Feng, J., Liu, T., Du, Y., Guo, S., Lin, Y., and Li, Y. (2025). Citygpt: Empowering urban spatial cognition of large language models. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2, KDD ’25, page 591–602, New York, NY, USA. Association for Computing Machinery.

He, J., Nie, T., and Ma, W. (2025). Geolocation representation from large language models are generic enhancers for spatio-temporal learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 17094–17104.

Liu, Z., Janowicz, K., Currier, K., and Shi, M. (2024). Measuring geographic diversity of foundation models with a natural language–based geo-guessing experiment on gpt-4.

Manvi, R., Khanna, S., Mai, G., Burke, M., Lobell, D., and Ermon, S. (2024). Geollm: Extracting geospatial knowledge from large language models. In Kim, B., Yue, Y., Chaudhuri, S., Fragkiadaki, K., Khan, M., and Sun, Y., editors, International Conference on Learning Representations, volume 2024, pages 38791–38807.

Moayeri, M., Tabassi, E., and Feizi, S. (2024). Worldbench: Quantifying geographic disparities in llm factual recall. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’24, page 1211–1228, New York, NY, USA. Association for Computing Machinery.

OECD (2026). Urban population by city size. [link]. Acesso em: 16 Abril 2026.

Yan, Y. and Lee, J. (2024). Georeasoner: Reasoning on geospatially grounded context for natural language understanding. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM ’24, page 4163–4167, New York, NY, USA. Association for Computing Machinery.