Skip to main content
. 2022 Mar 8;10:31513–31523. doi: 10.1109/ACCESS.2022.3157854

TABLE 2. Performance of Current BioNER Models on NCBI-Disease, BC5CDRdis, and BC5CDRchem. The Best Scores are Highlighted in Bold, and the Second Best Scores are Underlined.

Test
Training Model Overall In-depth COVID-19
P R F1 Inline graphic Inline graphic Inline graphic
NCBI-disease PubMedBERT [13] 86.6 88.9 87.7 94.5 81.1 77.7 36.0
BlueBERT [14] 85.8 89.5 87.6 95.9 80.4 76.7 13.8
BioBERT [7] 86.7 90.5 88.6 95.5 80.9 84.1 45.7
BERT [10] 83.8 87.6 85.6 95.0 79.0 70.7 18.7
DICTsyn 50.7 58.0 54.1 88.5 13.8 0.0 0.0
DICTtrain 52.7 55.4 54.0 88.8 0.0 0.0 0.0
BC5CDRdis PubMedBERT [13] 83.1 87.3 85.2 93.1 78.3 75.7 2.2
BlueBERT [14] 82.2 86.6 84.4 93.2 76.9 72.7 0.8
BioBERT [7] 82.4 86.3 84.3 93.3 74.9 73.7 3.4
BERT [10] 78.5 81.4 79.9 91.5 64.0 63.4 0.8
DICTsyn 75.4 67.8 71.4 96.0 32.9 0.0 0.0
DICTtrain 75.9 61.4 67.8 96.7 0.0 0.0 0.0
BC5CDRchem PubMedBERT [13] 92.1 94.2 93.1 98.3 85.5 88.2 -
BlueBERT [14] 92.8 92.9 92.8 98.0 81.6 86.0 -
BioBERT [7] 92.1 93.1 92.6 97.8 82.1 87.0 -
BERT [10] 89.8 90.0 89.9 97.0 72.4 81.1 -
DICTsyn 71.5 62.2 66.5 95.9 32.7 1.4 -
DICTtrain 71.2 58.8 64.6 96.2 0.0 0.0 -