NER Tag Schemes: IOB and BIOES
You'll learn: How to mark the boundaries of named entities in text using structured tagging schemes that tell models where entities begin, continue, and end.
Why Tag Schemes Matter
When performing Named Entity Recognition, you can't just label words with entity types like PERSON or LOCATION. Why? Because entities often span multiple words ("New York City"). You need a way to indicate:
- Which word starts an entity
- Which words continue it
- Which words are outside any entity
That's where tag schemes come in.
IOB Tagging (Inside-Outside-Beginning)
The IOB scheme uses three prefixes:
- B- (Beginning): First token of an entity
- I- (Inside): Continuation tokens of an entity
- O (Outside): Not part of any entity
Example:
"Barack Obama visited New York"
B-PER I-PER O B-LOC I-LOC
Notice how multi-word entities get B- for the first word, then I- for the rest.
BIOES Tagging (Begin-Inside-Outside-End-Single)
The BIOES scheme is more explicit with five tags:
- B- (Beginning): First token of a multi-token entity
- I- (Inside): Middle tokens
- E- (End): Last token of a multi-token entity
- S- (Single): Single-token entity
- O (Outside): Not part of any entity
Example:
"Barack Obama visited Paris"
B-PER E-PER O S-LOC
BIOES makes boundaries crystal clear—you know exactly where entities end, and distinguish single-word entities explicitly.
Key Takeaway: IOB and BIOES schemes convert entity recognition into a sequence labeling problem where each token gets a boundary-aware tag, enabling models to correctly identify multi-word entities.