deberta-v3-large trained for one epoch (properly underfitted) on a dataset that combined FEVER data with externally sourced non-claims. Label 0: no claim Label 1: claim
Paper of the data: "Claim extraction from text using transfer learning" - By Acharya Ashish Prabhakar, Salar Mohtaj, Sebastian Möller
https://aclanthology.org/2020.icon-main.39/
Performance on held out data:
Accuracy 0.8128525 F1_Score 0.854962 mcc 0.6173648
properly is of use already
- Downloads last month
- 17