A carregar...
Publicação
An exploratory study of using generative pre-trained transformer (GPT) models for geoparsing in a crisis crowdsourcing mapping workflow
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.87 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
User-generated messages with geographical information on social media
platforms has been found to play an increasingly important role in crisis
crowdsourced mapping. Manually extracting information requires significant
human effort and time, and traditional named entity recognition methods
have been shown to struggle with accurately extracting specific locations from
the messages. In this study, we propose an approach that combines
Generative Pre-trained Transformer (GPT) with Google geocoder to automate
geoparsing and geocoding without any training or additional data. The results
indicate that, compared to manual operations, the GPT-3.5 model achieves a
match rate of 55.15%, and the GPT-4 model achieves 62.28%. This suggests
that large language models, represented by the GPT models, have the
potential to be applied in crisis crowdsourced mapping, benefiting rapid
emergency responses and potentially saving lives.
Descrição
Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies
Palavras-chave
Crisis crowdsourcing mapping GPT Geoparsing Geocoding