Skip to main content
. 2023 May 12;39(5):btad310. doi: 10.1093/bioinformatics/btad310

Table 2.

F1 scores for multiple named entity recognition on the BioRED test set.a

Dataset Overall Gene Disease Chemical Species Variant CellLine
BioRED 89.34 92.35 83.47 88.55 96.98 87.34 90.53
+NLM-Gene 89.76 92.40 84.03 90.19 97.35 85.89 86.87
+GNormPlus 89.95 92.74 83.57 90.05 96.82 88.98 91.67
+NCBI-Disease 89.55 91.68 85.19 89.46 96.52 86.01 81.72
+BC5CDR-Disease 89.66 91.46 85.34 89.67 96.98 84.86 90.53
+BC5CDR-Chemical 89.40 91.52 84.07 89.09 96.99 88.38 87.50
+NLM-Chem 89.60 91.92 84.15 89.78 97.09 87.16 83.67
+Linnaeus 89.19 91.49 84.04 88.69 96.72 88.16 86.60
+Species-800 89.65 92.19 83.34 90.14 97.37 88.79 80.81
+tmVar3 89.01 91.08 83.77 88.09 97.08 89.21 88.66
+BioID 89.69 92.02 84.23 88.83 97.48 88.75 91.67
+All (w/o AIONER) 69.96 76.85 58.86 84.82 30.57 71.77 27.12
+All (MTL) 90.84b 92.59 87.01 90.71 96.40 88.25 90.32
+All (AIONER)

  • 91.26 b

  • (+1.92)

  • 92.40

  • (+0.05)

  • 88.07

  • (+4.60)

  • 90.98

  • (+2.43)

  • 97.50

  • (+0.52)

  • 88.51

  • (+1.17)

  • 90.53

  • (+0.00)
a

The parenthesized numbers are the improvements of AIONER compared to the baseline trained on the BioRED training set only. Bold indicates the best score for each entity type and overall entity.

b

P < 0.05 (two-sided Wilcoxon signed-rank test compared with baseline). There is no significant difference between MTL and AIONER.