Phenonizer: A Fine-Grained Phenotypic Named Entity Recognizer for Chinese Clinical Texts

Table 6

Comparison of symptom extraction performance for different models on the WithNeg dataset.

Models	Precision		Recall		F1-score
	Normal	Degraded	Normal	Degraded	Normal	Degraded
BiLSTM-CRF	0.9134	0.6111	0.9090	0.9133	0.9112	0.7322
GloVeWiki-BiLSTM-CRF	0.9077	0.6076	0.9085	0.9124	0.9081	0.7294
GloVeMedical-BiLSTM-CRF	0.9119	0.6120	0.9077	0.9127	0.9098	0.7327
W2VWiki-BiLSTM-CRF	0.9363	0.6133	0.9213	0.9287	0.9287	0.7387
W2VMedical-BiLSTM-CRF	0.9329	0.6144	0.9281	0.9321	0.9305	0.7406
BERT-CRF	0.9261	0.6211	0.9243	0.9254	0.9252	0.7433
Phenonizer	0.9405	0.6216	0.9387	0.9398	0.9396	0.7483