The contributions of this paper are as follows:
We provided the source code of uncertainty-aware unlikelihood learning (UAUL).
Overview of our Uncertainty-Aware Unlikelihood Learning method. We address "noise and errors" with an "acquire-and-reduce" method, using an uncertainty-aware negative sampling approach with Monte-Carlo Dropout (MC dropout) to obtain negative samples vulnerable to errors during training. We then propose a MUL loss to reduce these errors and noise.
The quadruple extraction performance of five different systems (including baseline and +UAUL) on the four datasets:
We further investigate the ability of UAUL under low resource scenario:
Python==3.8
torch==1.11.0
transformers==4.14.1
pytorch_lightning==0.8.1
numpy==1.21.2
If you use the data and code in your research, please cite our paper as follows: