Skip to main content

Table 3 Impact of Unary-Network and Pairwise-Network in terms of the F1-score (%)

From: DTranNER: biomedical named entity recognition with deep learning-based label-label transition model

Settings

BC5CDR-Chemical

BC5CDR-Disease

NCBI-Disease

Unary-CRF

93.01

86.14

86.94

Pairwise-CRF

93.27

86.05

86.71

Unary+Pairwise ensemble

93.25

86.78

87.09

DTranNER

94.16

87.22

88.62

  1. Note: “Unary-CRF” denotes a variant model excluding Pairwise-Network from DTranNER, “Pairwise-CRF” denotes a variant model excluding Unary-Network from DTranNER, and “Unary+Pairwise ensemble” is an ensemble model of “Unary-CRF” and “Pairwise-CRF.” In the ensemble model, “Unary-CRF” and “Pairwise-CRF” were independently trained, and they voted over the sequence predictions by their prediction scores