Customizing spaCy models

Natural Language Processing with spaCy

Azadeh Mobasher

Principal data scientist

Why train spaCy models?

Go a long way for general NLP use cases
But may not have seen specific domains data during their training, e.g.
- Twitter data
- Medical data

Example of a medical domain NER

Why train spaCy models?

Better results on your specific domain
Essential for domain specific text classification

Before start training, ask the following questions:

Do spaCy models perform well enough on our data?
Does our domain include many labels that are absent in spaCy models?

Models performance on our data

Do spaCy models perform well enough on our data?
Oxford Street is not correctly classified with a GPE label:

import spacy
nlp = spacy.load("en_core_web_sm")

text = "The car was navigating to the Oxford Street."
doc = nlp(text)
print([(ent.text, ent.label_) for ent in doc.ents])

[('the Oxford Street', 'ORG')]

Output labels in spaCy models

Does our domain include many labels that are absent in spaCy models?

NER example of common vs. medical domains

Output labels in spaCy models

If we need custom model training, we follow these steps:

Collect our domain specific data
Annotate our data
Determine to update an existing model or train a model from scratch

Let's practice!

Natural Language Processing with spaCy