terms-from-ner

Terms from NER Annotations

This folder contains a recipe that allows you to turn your pre-existing NER annotations into patterns that you can re-use. Having such patterns is great, because these can help pre-highlight text in Prodigy.

Try it out

To try this recipe, you can start by first annotating some data from news_headlines.jsonl (found in the example-datasets folder), via;

python -m prodigy ner.manual news_demo blank:en news_headlines.jsonl --label PERSON,ORG,PRODUCT,LOCATION

Then, after annotating a few examples, you can generate a patterns.jsonl file that contains patterns.

python -m prodigy terms.from-ner ner_news_headlines patterns.jsonl -F recipe.py

Finally, you can now re-use these patterns in your original ner.manual recipe, via:

python -m prodigy ner.manual news_demo blank:en news_headlines.jsonl --label PERSON,ORG,PRODUCT,LOCATION --patterns patterns.jsonl