-
Notifications
You must be signed in to change notification settings - Fork 20
Open
Description
Sometimes the spaCy NER step fails with very short documents, including examples where the entire document is just a single sentence, because it's lacking the usual context to identify locations. This happens especially with locations outside of Europe or English-speaking countries because of the limitations of the NER training data.
It would be nice to allow users to override the NER step and force a geolocation on a known entity.
The easiest way to do this would probably be to:
- pull most of the geolocation logic out of
geoparse_doc
into a separate function - create a new function (
geoparse_ent
?) that takes a document + entity string as arguments. It can then manually add the string entity as a spaCy entity, and then call the new function above.
Metadata
Metadata
Assignees
Labels
No labels