TopoBERT: a plug and play toponym recognition module harnessing fine-tuned BERT

Extracting precise geographical information from the textual content, referred to as toponym recognition, is fundamental in retrieval and crucial a plethora of spatial analyses, e.g. mining location-based social media, news reports, surveys for various applications. However, performance existing recognition methods tools deficient supporting tasks that rely on extracting fine-grained geographic texts, locating people sending help requests with addresses through media during disasters. The emerging pretrained language models have revolutionized natural processing understanding by machines, offering promising pathway optimize underpin practical In this paper, TopoBERT, uniquely designed module based one-dimensional Convolutional Neural Network (CNN1D) Bidirectional Encoder Representation Transformers (BERT), proposed fine-tuned. Three datasets are leveraged tune hyperparameters discover best strategy train model. Another seven used evaluate performance. TopoBERT achieves state-of-the-art (average f1-score = 0.854) compared baseline models. It encapsulated into easy-to-use python scripts can be seamlessly applied diverse without additional training.