@InProceedings{miyazaki-EtAl:2018:W-NUT2018,
  author    = {Miyazaki, Taro  and  Rahimi, Afshin  and  Cohn, Trevor  and  Baldwin, Timothy},
  title     = {Twitter Geolocation using Knowledge-Based Methods},
  booktitle = {Proceedings of the 2018 EMNLP Workshop W-NUT: The 4th Workshop on Noisy User-generated Text},
  month     = {November},
  year      = {2018},
  address   = {Brussels, Belgium},
  publisher = {Association for Computational Linguistics},
  pages     = {7--16},
  abstract  = {Geolocation of user posts on Twitter is useful for many applications, including disaster monitoring and news material gathering. However, the vast majority of tweets have no explicit geotag, motivating the need for automatic geolocation prediction methods. We propose the use of named entity linking in geolocation prediction, modelled using graph convolutional networks over a knowledge base of entity relations, which is combined with text-based models in an end-to-end deep learning framework. We show that our method improves on text-based models, and learns effective representations for named entities that do not appear in the training data.},
  url       = {http://www.aclweb.org/anthology/W18-6102}
}