Skip to main content

Table 2 Test results in biomedical NER task using the micro-averaged F1-score

From: Entity and relation extraction from clinical case reports of COVID-19: a natural language processing approach

Model/Dataset

NCBI

BC5 CDR

BC4 CHEMD

BC2 GM

JNL PBA

i2b2-clinical

i2b2-2012

Our data

BiLSTM-based

BiLSTM-CRF [63]

85.80

85.92

89.48

80.60

74.29

85.66

86.10

88.10

BILSTM-CNN-Char [71]

89.73

87.73

88.06

87.51

77.29

84.08

83.10

89.20

BiLSTM-CRF-MTL [72]

88.85

84.93

89.42

82.12

77.03

83.25

83.24

87.10

Att-BiLSTM-CRF [73]

84.50

88.93

91.08

84.37

77.10

88.60

85.83

88.10

Doc-Att-BiLSTM-CRF [16]

88.60

87.30

85.10

81.80

76.23

85.18

85.17

86.94

BiLSTM-contextualized [74]

85.17

87.83

87.12

80.51

73.52

85.26

84.22

85.18

CollaboNet [75]

84.08

84.08

87.12

79.73

78.58

85.61

84.29

86.70

BERT-based

SciBERT [76]

86.88

87.94

88.06

84.08

75.77

79.19

79.10

81.95

BLUE [77]

86.37

86.60

90.19

80.93

75.27

84.06

83.45

89.95

BioBERT-Base v1.0 [13]

87.71

85.80

90.77

84.72

77.59

85.64

86.00

90.10

BioBERT-Base v1.1 [13]

88.30

84.67

90.78

88.41

77.21

86.73

86.23

90.39

BioBERT-Base v1.2 [13]

88.15

86.70

92.20

88.72

79.10

88.95

87.40

92.34

Our approach

90.58

89.90

91.58

89.15

79.92

89.10

90.10

94.78

  1. Bold means best score, italic second-best score