Skip to main content

Table 11 Detailed results achieved by using different scope of feature channels of MSTLTR. The following situations are included: no shared features are used; only global shared features from 1 to 4 source domain datasets are used; both global common and local common features from 4 source domain datasets are used

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

Models Precision Recall F1-measure
Model I: no common feature (Basic Model) 79.47 77.23 78.34
Model II-1: global common features (Single Source) 82.25 77.89 80.01
Model II-2: global common features (Two Sources) 82.44 79.10 80.74
Model II-3: global common features (Three Sources) 82.75 78.35 80.49
Model II-4: global common features (Four Sources) 82.40 77.71 79.99
Model III: Model II-4 + local common features (MSTLTR) 83.96 79.89 81.88