LLM Agents for Georelating - A New Task for Locating Events
Abstract: Accurately identifying disaster-affected areas is crucial for data-driven disaster resilience. In response, we introduce Georelating, a task that infers affected areas from textual reports containing complex locative expressions, moving beyond traditional geoparsing approaches that rely on explicit point locations. Georelating instead combines resolving unnamed regions and reasoning about spatial relations to represent event-affected areas within standardized Discrete Global Grid Systems (DGGSs). We propose addressing Georelating with a pipeline capitalizing on the contextual understanding of large language model (LLM) agents to perform geospatial reasoning. Preliminary evaluation highlights the potential of this approach for the foundational geocoding stage and the novel Georelating task. We point out future paths for enhancing Georelating systems toward intuitive and efficient disaster information systems.
Show BibTeX
@inproceedings{DBLP:conf/gis/MoltzenHU25,
author = {Kai Moltzen and
Junbo Huang and
Ricardo Usbeck},
editor = {Mohamed F. Mokbel and
Shashi Shekar and
Andreas Z{\"{u}}fle and
Yao{-}Yi Chiang and
Maria Luisa Damiani and
Moustafa A. Youssef},
title = {{LLM} Agents for Georelating - {A} New Task for Locating Events},
booktitle = {Proceedings of the 33rd {ACM} International Conference on Advances
in Geographic Information Systems, {SIGSPATIAL} 2025, The Graduate
Hotel Minneapolis, Minneapolis, MN, USA, November 3-6, 2025},
pages = {277--280},
publisher = {{ACM}},
year = {2025},
url = {https://doi.org/10.1145/3748636.3762733},
doi = {10.1145/3748636.3762733},
timestamp = {Tue, 03 Feb 2026 08:26:58 +0100},
biburl = {https://dblp.org/rec/conf/gis/MoltzenHU25.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}