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

Table 9

The symptom extraction performance of models on heterogenous data (TCM-HB).

Training dataset	TCM-HB			TCM-HN
Models	Precision	Recall	F1-score	Precision	Recall	F1-score
BiLSTM-CRF	0.7682	0.7865	0.7772	0.6512	0.5865	0.6171
GloVeWiki-BiLSTM-CRF	0.7701	0.7870	0.7785	0.6510	0.6097	0.6297
GloVeMedical-BiLSTM-CRF	0.7705	0.7957	0.7829	0.6575	0.6104	0.6331
W2VWiki-BiLSTM-CRF	0.7686	0.7964	0.7822	0.6436	0.6261	0.6347
W2VMedical-BiLSTM-CRF	0.7734	0.7996	0.7863	0.6623	0.6139	0.6372
BERT-CRF	0.7719	0.8179	0.7943	0.6566	0.6198	0.6377
BERT-BiLSTM	0.7688	0.8145	0.7910	0.6400	0.6406	0.6403
Phenonizer	0.7727	0.8189	0.7952	0.6438	0.6446	0.6442