Models

F1

4 way (Accuracy)

3 way (Accuracy)

Binary (Accuracy)

BERT-single

91.06%

82.92%

86.12%

92.83%

BERT-pair QA_M

91.66%

84.19%

88.48%

93.74%

BERT-pair NLI_M

92.06%

84.39%

87.66%

93.85%

BERT-pair QA_B

92.35%

85.36%

89.41%

94.31%

BERT-pair NLI_B

91.56%

84.4%

88.0%

94.7%