Transformer Based Geocoding

TL;DR

This paper introduces a transformer-based sequence-to-sequence approach for geocoding, predicting geolocations from free text using a T5 model trained on geo-tagged data, with publicly available code and datasets.

Contribution

It formulates geocoding as a sequence-to-sequence task and trains a T5 transformer model specifically for this purpose, which is a novel application of transformers in geocoding.

Findings

Effective geolocation prediction from free text.

Open-source code and datasets available for reproducibility.

Adaptive cell partitioning improves geolocation accuracy.

Abstract

In this paper, we formulate the problem of predicting a geolocation from free text as a sequence-to-sequence problem. Using this formulation, we obtain a geocoding model by training a T5 encoder-decoder transformer model using free text as an input and geolocation as an output. The geocoding model was trained on geo-tagged wikidump data with adaptive cell partitioning for the geolocation representation. All of the code including Rest-based application, dataset and model checkpoints used in this work are publicly available.