Domain-Specific Language Model Pretraining for Biomedical Natural Language Processing

Benchmark Model Rank Results
participant-intervention-comparison-outcome-extraction-on-ebm-nlpPubMedBERT uncased#1F1: 73.38
drug-drug-interaction-extraction-on-ddiPubMedBERT#3F1: 0.8236Micro F1: 82.36
pico-on-ebm-picoPubMedBERT uncased#3Macro F1 word level: 73.38
question-answering-on-blurbPubMedBERT (uncased; abstracts)#3Accuracy: 71.7
relation-extraction-on-ddiPubMedBERT uncased#3Micro F1: 82.36
relation-extraction-on-gadPubMedBERT uncased#3Micro F1: 82.34
text-classification-on-blurbPubMedBERT (uncased; abstracts)#3F1: 82.32
document-classification-on-hocPubMedBERT uncased#5Micro F1: 82.32
question-answering-on-bioasqPubMedBERT uncased#5Accuracy: 87.56
named-entity-recognition-ner-on-jnlpbaPubMedBERT uncased#9F1: 79.1
named-entity-recognition-on-bc2gmPubMedBERT uncased#10F1: 84.52
relation-extraction-on-chemprotPubMedBERT uncased#12Micro F1: 77.24
named-entity-recognition-ner-on-ncbi-diseasePubMedBERT uncased#16F1: 87.82
question-answering-on-pubmedqaPubMedBERT uncased#32Accuracy: 55.84