Skip to main content

Table 2 Experimental results of models on six benchmark datasets

From: Improving biomedical named entity recognition with syntactic information

Methods

BC2GM

JNLPBA

BC5CDR-chemical

NCBI-disease

LINNAEUS

Species-800

F1

\(\sigma\)

F1

\(\sigma\)

F1

\(\sigma\)

F1

\(\sigma\)

F1

\(\sigma\)

F1

\(\sigma\)

Base

84.61

0.21

76.85

0.31

93.50

0.10

88.63

0.71

88.27

0.32

74.97

0.46

+ PL (DC)

84.47

0.15

77.17

0.45

93.66

0.15

89.09

0.55

88.36

0.16

75.04

0.46

+ PL (\({\mathcal {M}}\))

\(\mathit{84} .\mathit{74}\)

0.10

77.06

0.05

\(\mathit{93} .\mathit{73}\)

0.19

\(\mathit{89} .\mathit{47}\)

0.56

\(\mathit{88} .\mathit{44}\)

0.30

\(\mathit{75} .\mathit{45}\)

0.41

+ SC (DC)

84.45

0.19

76.80

0.45

93.68

0.13

89.18

0.26

88.23

0.33

75.37

0.51

+ SC (\({\mathcal {M}}\))

\(\mathit{84} .\mathit{76}\)

0.21

\(\mathit{77} .\mathit{17}\)

0.16

\(\mathit{93} .\mathit{74}\)

0.11

\(\mathit{89} .\mathit{27}\)

0.52

\(\mathit{88} .\mathit{68}\)

0.30

\(\mathit{75} .\mathit{65}\)

0.50

+ DR (DC)

84.33

0.30

77.01

0.28

93.66

0.15

89.05

0.23

88.43

0.19

75.12

0.52

+ DR (\({\mathcal {M}}\))

\(\mathit{84} .\mathit{65}\)

0.27

\(\mathit{77} .\mathit{32}\)

0.35

\(\mathit{93} .\mathit{78}\)

0.18

\(\mathit{89} .\mathit{24}\)

0.60

\(\mathit{88} .\mathit{57}\)

0.15

\(\mathit{75} .\mathit{81}\)

0.71

Large

84.89

0.17

77.29

0.19

93.90

0.31

88.65

0.59

88.87

0.65

74.98

0.59

+ PL (DC)

85.06

0.08

\(\mathit{77} .\mathit{56}\)

0.18

93.90

0.16

88.74

0.26

88.65

0.39

74.92

0.86

+ PL (\({\mathcal {M}}\))

\(\mathit{85} .\mathit{07}\)

0.12

77.50

0.19

\(\mathit{94} .\mathit{05}\)

0.23

\(\mathit{88} .\mathit{86}\)

0.29

\(\mathit{89} .\mathit{01}\)

0.31

\(\mathit{75} .\mathit{34}\)

0.95

+ SC (DC)

85.12

0.13

77.56

0.12

93.95

0.09

88.78

0.54

\(\mathit{89} .\mathit{01}\)

0.28

\(\mathit{75} .\mathit{38}\)

0.29

+ SC (\({\mathcal {M}}\))

\(\mathit{85} .\mathit{43}\)

0.15

\(\mathit{77} .\mathit{83}\)

0.19

\(\mathit{93} .\mathit{99}\)

0.13

\(\mathit{88} .\mathit{87}\)

0.37

88.92

0.35

75.08

0.68

+ DR (DC)

85.01

0.12

77.58

0.10

93.97

0.17

\(\mathit{89} .\mathit{37}\)

0.30

88.99

0.22

75.01

0.83

+ DR (\({\mathcal {M}}\))

\(\mathit{85} .\mathit{17}\)

0.10

\(\mathit{77} .\mathit{73}\)

0.11

\(\mathit{94} .\mathit{05}\)

0.10

88.81

0.51

\(\mathit{89} .\mathit{04}\)

0.27

\(\mathit{75} .\mathit{17}\)

0.91

  1. The experimental results are reported in terms of average F1 scores (F1) and the standard deviation \(\sigma\). The methods in the group “Base” and “Large” refer to baselines with BioBERT-Base and BioBERT-Large encoder and our methods with KVMN (\({\mathcal {M}}\)). “DC” refers to the baseline method using direct concatenation to incorporate syntactic information. “PL”, “SC”, and “DR” stand for POS labels, syntactic constituents, and dependency relations, respectively