Named Entity Recognition (NER) using LSTM with Keras

Dataset Source

• NER Dataset

Model Used

1. Bi-directional LSTM
2. Bi-directional GRU

Notes: Bi-directional GRU performed well.

Accuracy

model	training loss	training acc	validation loss	validation acc
Bi-directional LSTM	0.039	0.988	0.048	0.985
Bi-directional GRU	0.034	0.989	0.045	0.986

Performance (Bi-directional GRU)

Some prediction (Bi-directional GRU)

get_ner('Sunday is close day for London .')

Sunday         	B-tim
London         	B-geo

get_ner('I will go to London from Indonesia Sunday Morning .')

London         	B-geo
Indonesia      	B-geo
Sunday         	B-tim
Morning        	I-tim

get_ner('U.S. President George Bush will visit the Vatican city Tomorrow Morning .')

U.S.           	B-org
President      	B-per
George         	I-per
Bush           	I-per
Vatican        	B-org
Tomorrow       	B-geo
Morning        	I-org