Disaster Tweets, Geocoded Locations

Context

Trying to make use of the location feature in the "Real or Not? NLP with Disaster Tweets" competition.
I tried to geocode the locations, hoping that at least the difference between locations that can be geocoded (e.g. Birmingham) vs those that cannot be (e.g. "your sisters bedroom") would be a good feature. Additionally, geocoding provides longitude and latitude features that may be helpful.

Content

The dataset captures whether a location could be geocoded (that is: it is a valid location in the world).

Acknowledgements

Geocoding is done with Nominatim

Inspiration

Can you make better tweet classifications with geocoded locations?