CPLL

This is the code for our paper: A Confidence-based Partial Label Learning Model for Crowd-Annotated Named Entity Recognition (ACL 2023)

Dependencies

pip install transformers

Reproduction

Preprocess Data

First put the data (train.txt, dev.txt, test.txt, labels.txt) in a folder. The data format refers to the CoNLL 2003 dataset.

Then to simulate a crowd-annotated situation, you can run our perturb script.

python preprocess.py --raw_data_dir raw_data_dir --output_dir output_dir

Prepare a pre-trained model

We used Chinese-roberta-wwm-ext downloaded from here

Training

Here is an example

python run.py --bert_dir pre-trained_model_dir --data_dir data_dir --ent2id_dir data_dir

Citation

If you use our results or scripts in your research, please cite our paper.

@article{xiong2023confidence,
  title={A Confidence-based Partial Label Learning Model for Crowd-Annotated Named Entity Recognition},
  author={Xiong, Limao and Zhou, Jie and Zhu, Qunxi and Wang, Xiao and Wu, Yuanbin and Zhang, Qi and Gui, Tao and Huang, Xuanjing and Ma, Jin and Shan, Ying},
  journal={arXiv preprint arXiv:2305.12485},
  year={2023}
}