| Title: |
Geospatial Data Quality in the Era of Generative AI: Can we Trust the Geographic Information Produced by Large Language Models? |
| Authors: |
Shingleton, Joseph; Basiri, Ana |
| Publication Year: |
2025 |
| Collection: |
University of Glasgow: Enlighten - Publications |
| Subject Terms: |
QA76 Computer software |
| Description: |
With the rising interest and experimentation of generative artificial intelligence (GenAI) across many fields and multiple applications, it is becoming increasingly important that we are able to properly assess the quality of the data provided by the models, particularly in light of challenges such as hallucinations and other data accuracy issues that can undermine their reliability. For location based services that build upon GenAI outputs, geospatial data quality is essential and feasible because we deal with real-world features, i.e. the literal ground truth. In this paper, we present two experiments which probe the quality of the geospatial information provided by large language models. The first experiment asks a model to geolocate place names in a piece of text multiple times, highlighting both geospatial accuracy and precision. In the second, we ask the model to plan routes using the London Underground network, investigating the models’ propensity for temporal and topological inconsistencies and thematic inaccuracies. The results of these experiments highlight the need for responsible practices when deploying GenAI in geospatial applications. |
| Document Type: |
conference object |
| File Description: |
text |
| Language: |
English |
| Relation: |
https://eprints.gla.ac.uk/349410/2/349410.pdf; Shingleton, Joseph ORCID logoorcid:0000-0002-1628-3231 and Basiri, Ana ORCID logoorcid:0000-0002-2399-1797 (2025) Geospatial Data Quality in the Era of Generative AI: Can we Trust the Geographic Information Produced by Large Language Models? 19th International Conference on Location Based Services (LBS 2025), Otaniemi, Espoo, Finland, 7-9 May 2025. (Unpublished) |
| Availability: |
https://eprints.gla.ac.uk/349410/; https://eprints.gla.ac.uk/349410/2/349410.pdf |
| Accession Number: |
edsbas.DF3299A3 |
| Database: |
BASE |