Leveraging a Joint learning Model to Extract Mixture Symptom Mentions from Traditional Chinese Medicine Clinical Notes

Table 4

Entity recognition and relation extraction with pipeline approaches.

Model	Label embedding	F1-score	Precision	Recall
Relation extraction pipeline	Without	0.6794	0.8374	0.5716
Multihead joint learning	Without	0.7079	0.8228	0.6212
BERT+relation extraction pipeline	Without	0.7222	0.7596	0.6884
BERT+relation extraction pipeline	With	0.7851	0.8496	0.7297
BERT+multihead joint learning	Without	0.8016	0.8218	0.7823
BERT+multihead joint learning	With	0.8216	0.8250	0.8183