Custom NER Tag for english


How to train a custom NER tag using prodigy/spacy


CSK won a IPL Cricket cup 2018. Dhoni is the best captain so far.

Dhoni - Person

Kindly help us

Hi! A good place to start is the ner.teach recipe, which takes a pre-trained spaCy model like en_core_web_sm, an input text and a list of labels, e.g. --label PERSON,ORG. Prodigy will then show you suggestions of entities, and you can accept or reject them, and correct the model's predictions on your data so it makes fewer mistakes.

If you're new to spaCy and haven't seen it already, a good place to start is the documentation on training models. It explains what's going on under the hood, and how the entity recognizer is updated:

This Prodigy tutorial shows an end-to-end example of improving spaCy's existing entity types on new data. The example trains PRODUCT, but you can obviously use the same approach for PERSON and ORG:

You might also find our NER video tutorial useful. It shows how to train a new category from scratch, so the use case is a little different – but it shows the different steps of working with Prodigy, the available commands and explains the philosophy behind the application design:

Finally, the PRODIGY_README.html that's available for download with Prodigy includes the full recipe and API docs. You can also browse the ner tag on this forum for more examples, solutions and best practices.