BIO tagging
label each token with B-TYPE (begin), I-TYPE (inside), or O (outside)
token: Apple sued Google over its iPhone search deal in the US .
label: B-ORG O B-ORG O O B-PROD O O O O B-GPE O
rule-based
gazetteer match
regex + dicts
high precision
zero OOV coverage
no disambiguation
CRF
hand features:
shape, suffix,
neighbors, case
enforces valid
BIO transitions
BiLSTM-CRF
learned features
embed → BiLSTM
→ CRF head
no more hand
engineering
transformer NER
fine-tune BERT
token-classif head
aggregate spans
best accuracy
most compute
pick the cheapest method that hits your accuracy bar