Table 3.

The micro-average performances for concept extraction and relation identification.a

Models	Subtask 1 (concept extraction)			Subtask 2 (relation identification)
	Precision	Recall	F1 score		Precision	Recall	F1 score
LSTMa-CRFsb + BERTc-cls + BERT-rel	0.7760	0.8087	0.7920		0.7343	0.5465	0.6266
LSTM-CRFs-EN + BERT-cls + BERT-reld	0.7969	0.7920	0.7944		0.6995	0.6184	0.6544
BERT-ner + BERT-cls + BERT-rel	0.8060	0.8105	0.8083		0.7140	0.6252e	0.6667
BERT-ner-EN + BERT-cls + BERT-rel	0.8301e	0.8198e	0.8249e		0.7421e	0.6233	0.6775e

^aLSTM: long short-term memory.

^bCRFs: conditional random fields.

^cBERT: bidirectional encoder representations from transformers.

^dOur best system developed during the challenge.

^eThe best performances.