Skip to main content

Table 6 Impact of dataset size

From: Multitask learning for biomedical named entity recognition with cross-sharing structure

   Full-size STM Full-size CS-MTM 50%-size STM 50%-size CS-MTM 25%-size STM 25%-size CS-MTM 10%-size STM 10%-size CS-MTM
BC2GM Precision 84.00 83.12 82.37 79.37 77.82 79.44 73.19 72.95
  Recall 83.82 85.74 80.77 85.05 79.57 78.98 73.59 75.39
  F1 83.91 84.41 81.56 82.12 78.69 79.21 73.39 74.15
Ex-PTM Precision 70.83 74.73 67.74 68.18 57.46 54.00 42.47 50.69
  Recall 64.12 69.56 58.62 67.48 53.69 63.97 50.27 41.68
  F1 67.31 72.05 62.85 67.83 55.51 58.56 46.04 45.75
NCBI-disease Precision 88.45 86.59 84.03 84.72 81.52 81.00 81.02 79.32
  Recall 83.78 86.42 84.56 84.76 76.50 81.00 68.59 74.40
  F1 86.05 86.50 84.30 84.74 78.93 81.00 74.29 76.78
Linnaeus Precision 92.86 89.81 91.77 88.92 89.90 90.20 90.80 85.98
  Recall 67.62 76.12 68.11 72.95 67.62 68.29 52.65 51.33
  F1 78.25 82.40 78.19 80.15 77.18 77.73 66.65 64.29
  1. Bold: the better F1 scores between STM and CS-MTM for each dataset size