This repository contains the code for our paper E-NER: Evidential Deep Learning for Trustworthy Named Entity Recognition (ACL Findings, 2023).
We study the problem of trustworthy NER by leveraging evidential deep learning. To address the issues of sparse entities and OOV/OOD entities, we propose E-NER with two uncertainty-guided loss terms. The uncertainty estimation quality of E-NER is improved without harming performance. Additionally, the well-qualified uncertainties contribute to detecting OOV/OOD, generalization, and sample selection.
Run the following script to install the dependencies,
pip3 install -r requirements.txt
The download links of the datasets used in this work are shown as follows:
For SpanNER and BERT-Tagger, we use BERT-Large. For Seq2Seq Model (https://github.com/yhcc/BARTNER), we use BART-Large
If you find this model helpful, please consider citing the following related papers.