Skip to main content

Table 3 Model performance comparison on the five benchmark datasets

From: A prefix and attention map discrimination fusion guided attention for biomedical named entity recognition

Model

BC4CHEMD

BC2GM

BC5CDR-Disease

BC5CDR-Chem

NCBI-Disease

\(\textbf{P}\)

\(\textbf{R}\)

\(\textbf{F1}\)

\(\textbf{P}\)

\(\textbf{R}\)

\(\textbf{F1}\)

\(\textbf{P}\)

\(\textbf{R}\)

\(\textbf{F1}\)

\(\textbf{P}\)

\(\textbf{R}\)

\(\textbf{F1}\)

\(\textbf{P}\)

\(\textbf{R}\)

\(\textbf{F1}\)

Habibi et al. [16]

–

–

–

81.57

79.48

80.51

–

–

–

87.60

86.25

86.92

86.11

85.49

85.80

Luo et al. [17]

–

–

–

–

–

–

89.61

83.09

86.23

–

–

–

90.72

74.89

82.05

Sachan et al. [18]

–

–

–

81.81

81.57

81.69

–

–

–

–

–

–

86.41

88.31

87.34

Devlin et al. [19]

–

–

–

81.17

82.42

81.79

–

–

–

90.94

91.38

91.16

84.12

87.19

85.63

Lee et al. [20]

92.80

91.92

92.36

84.32

85.12

84.72

86.47

89.84

87.15

93.68

93.26

93.47

88.22

91.25

89.71

Yu et al. [21]

–

–

–

–

–

84.52

–

–

85.62

–

–

93.33

–

–

87.82

Tian et al. [10]

–

–

–

–

–

84.92

–

–

–

–

–

94.00

–

–

90.08

Wang et al. [22]

90.78

87.53

89.37

82.10

79.42

80.74

–

–

–

–

–

–

85.86

86.42

86.14

Yoon et al. [23]

–

–

–

80.49

78.99

79.93

–

–

–

94.26

92.38

93.31

85.48

87.27

86.36

Khan et al. [24]

–

–

–

82.10

84.04

83.01

–

–

–

88.46

90.52

89.48

86.73

89.70

88.19

Chai et al. [25]

–

–

92.42*

–

–

82.92

–

–

87.28*

–

–

93.83

–

–

89.25

Tong et al. [26]

–

–

–

84.42

85.14

84.78*

–

–

–

93.29

94.69

93.98*

88.90

90.04

89.91*

PAMDFGA

91.74

93.37

92.55

85.43

85.47

85.45

87.11

87.95

87.53

93.66

94.67

94.16

89.76

91.35

90.55

  1. We separate the comparison methods of different categories with horizontal lines
  2. * indicates the best result without using additional knowledge. _ denotes the best result integrating external knowledge
  3. P is the Precision, R is the Recall, and F1 means the F1-score
  4. Data are expressed in percentage signs (\(\%\)). Bold F1-score represent our results