Result Analysis

Macro Precision, Recall, F1 (Word2Vec)

Micro Precision, Recall, F1 (Word2Vec)

Macro Precision, Recall, F1 (GloVe)

Micro Precision, Recall, F1 (GloVe)

Macro Precision, Recall, F1 (fastText)

Micro Precision, Recall, F1 (fastText)

•Dropout can improve accuracy

•BGRU based models have better accuracy than BLSTM based models

•CRF with BGRU and CNN have superior efficiency

•Word + Character Embedding > Word Embedding

•Word2Vec > GloVe, fastText

Winner: BGRU + CNN + CRF

•Summary of the winner model

•Confusion Matrix of the winner model