Skip to main content
. 2022 Jul 6;23:263. doi: 10.1186/s12859-022-04810-y

Table 7.

Entity-level results of BiLSTM-CRF models.

Label Precision Recall F1 Support
Random initialization
DISEASE 0.5414 0.3780 0.4451 454
RAREDISEASE 0.6540 0.7144 0.6829 1095
SIGN 0.4892 0.4391 0.4628 958
SYMPTOM 0.8529 0.5800 0.6905 54
Micro-avg 0.5421 0.5494 0.5457 2561
Macro-avg 0.5075 0.4223 0.4563 2561
Macro-weighted 0.5748 0.5494 0.5582 2561
Google news
DISEASE 0.5597 0.4304 0.4866 454
RAREDISEASE 0.6482 0.7548 0.6975 1095
SIGN 0.5327 0.4166 0.4675 958
SYMPTOM 0.6667 0.5600 0.6087 54
Micro-avg 0.5556 0.5654 0.5604 2561
Macro-avg 0.4815 0.4324 0.4521 2561
Macro-weighted 0.5887 0.5654 0.5711 2561
Glove
DISEASE 0.4720 0.5092 0.4899 454
RAREDISEASE 0.7226 0.7240 0.7233 1095
SIGN 0.5068 0.4606 0.4826 958
SYMPTOM 0.5385 0.5600 0.5490 54
micro-avg 0.5489 0.5821 0.5650 2561
Macro-avg 0.4480 0.4508 0.4490 2561
Macro-weighted 0.5937 0.5821 0.5874 2561
Wiki-pubmed-PMC
DISEASE 0.7208 0.4890 0.5827 454
RAREDISEASE 0.6339 0.7890 0.7030 1095
SIGN 0.4994 0.4562 0.4768 958
SYMPTOM 0.6739 0.5741 0.6200 54
Micro-avg 0.5564 0.6068 0.5805 2561
Macro-avg 0.5056 0.4617 0.4765 2561
Macro-weighted 0.5998 0.6068 0.5953 2561

Best micro and macro scores are in bold. Best scores for each entity type are also in bold