Named Entity Recognition

Motivating Named Entity Recognition

• An EntityRecognizer object will assign named entity labels to a Span of tokens
• A Span object is a slice from a Doc object
• A named entity is a real-world object that is assigned a name
• These names typically refer to a person, country, product, book title, etc.
• Spacy is able to recognize various types of named entities
• An EntityRecognizer uses a CNN model to predict these named entities

Accessing Entity Annotations

• The standard way of accessing entity annotations is the Doc.ents property
• This will produce a sequence of Span objects
• The Span object acts as a sequence of tokens, so we can iterate over the entity
• We can also retrieve information entity annotations for each Token using an IOB scheme

• The IOB scheme includes the following annotations:

  • I: Token is inside an entity
  • O: Token is outside an entity
  • B: Token is the beginning of an entity

Sample Code

>>> import spacy
>>> from spacy.tokens import Span

>>> doc = nlp("San Francisco considers banning sidewalk delivery robots")

# Document level entity annotations
>>> [(e.text, e.start_char, e.end_char, e.label_) for e in doc.ents]
[('San Francisco', 0, 13, 'GPE')]

# Token level entity annotations
>>> [doc[0].text, doc[0].ent_iob_, doc[0].ent_type_]
['San', 'B', 'GPE']
>>> [doc[1].text, doc[1].ent_iob_, doc[1].ent_type_]
['Francisco', 'I', 'GPE']

# Create new Doc
>>> doc = nlp("fb is hiring a new vice president of global policy")

# Check if organization
>>> [(e.text, e.start_char, e.end_char, e.label_) for e in doc.ents]
[]

# Set entity annotation
>>> fb_ent = Span(doc, 0, 1, label="ORG") # create Span for new entity
>>> doc.ents = list(doc.ents) + [fb_ent]
>>> [(e.text, e.start_char, e.end_char, e.label_) for e in doc.ents]
[('fb', 0, 2, 'ORG')]

References

• Spacy Named Entities
• EntityRecognizer Class
• Named Entity Annotations
• Span Class