Bi-Directional GRU CRF for Named Entity Recognition.
Dataset Source: https://www.kaggle.com/abhinavwalia95/entity-annotated-corpus (CoNLL 2002)
(only using first 10,000 samples with sentence length<=20 for training, validation, and testing)
The paper introducing CRF (Conditional Random Field):
Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data - by John Lafferty, Andrew McCallum, Fernando C.N. Pereira
CRF in details: An Introduction to Conditional Random Fields - by Charles Sutton and Andrew McCallum
Notes I used to understand CRF:
http://pages.cs.wisc.edu/~jerryzhu/cs838/CRF.pdf
https://guillaumegenthial.github.io/sequence-tagging-with-tensorflow.html
https://www.youtube.com/user/hugolarochelle/search?query=crf
This model is somewhat based on:
Bidirectional LSTM-CRF Models for Sequence Tagging - by Zhiheng Huang, Wei Xu, Kai Yu