Skip to main content

Table 9 Detailed results achieved by the proposed MSTLTR Model with different number of source datasets. All source datasets are numbered from S1 to S4 in the order of DataST09,DataEPI11,DataDNAm and DataDI. In the first “No Source” mode, the trigger recognition result without transfer learning is displayed. Then in the second “One source” mode, all the results using only one source dataset are listed. In the third “Two sources” mode, only the results of the combinations of the best single source dataset (S1) and other datasets are listed. Finally, “Multi-source” mode shows the results of multiple source domain transfer learning, including 3 and 4 source datasets. The illustrated 3 source dataset results are obtained based on the best “Two Sources” results. In each mode, the average results of all possible combinations of the source domains are listed by “AVG”

From: A transfer learning model with multi-source domains for biomedical event trigger extraction

Mode

Source domain

Precision

Recall

F1-measure

No Source

- (Basic Model)

79.47

77.23

78.34

One Source

DataST09(S1)

82.25

77.89

80.01

 

DataEPI11(S2)

81.74

76.60

79.09

 

DataDNAm(S3)

81.99

76.44

79.12

 

DataDI(S4)

82.80

77.24

79.92

 

AVG

-

-

79.53

Two Sources

S1+S2

81.79

78.78

80.26

 

S1+S3

83.28

77.77

80.43

 

S1+S4

83.16

78.56

80.80

 

AVG

-

-

80.24

Multi-Source

S1+S2+S4

83.62

78.36

80.90

 

S1+S3+S4

84.10

78.36

81.13

 

S1+S2+S3+S4(MSTLTR Model)

83.96

79.89

81.88

 

AVG

-

-

81.25