Skip to main content

Table 6 Cross-corpora evaluation by continual learning (CL), multi-corpora learning (ML), and continual multi-corpora learning (CML), single-head classifier (S), and multi-head classifier (M)

From: JCBIE: a joint continual learning neural network for biomedical information extraction

Training methods

CLS

ADE

DDI- ADE

DDI- CPR- ADE

ADE- DDI

ADE- DDI- CPR

Avg.

CL

S

29.99

33.69

31.95

1.04

0.01

19.34

M

30.06

34.13

32.04

26.73

23.21

29.23

ML

S

29.41

28.88

25.62

28.88

25.62

27.68

M

30.76

32.29

26.66

32.29

26.66

29.73

CML

S

25.21

36.46

27.08

20.13

22.51

26.28

M

28.16

38.95

33.83

28.01

29.18

31.63

Avg.

S

28.20

33.01

28.22

16.68

16.05

24.43

M

29.66

35.12

30.84

29.01

26.35

30.20

  1. The bold means the best results
  2. The performance is measured by RE micro-F1 on ADE\(_{3}\) corpus