DiffusionNER

Code for DiffusionNER: Boundary Diffusion for Named Entity Recognition, Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, Yueting Zhuang, ACL 2023.

During training, DiffusionNER gradually adds noises to the golden entity boundaries by a fixed forward diffusion process and learns a reverse diffusion process to recover the entity boundaries. In inference, DiffusionNER first randomly samples some noisy spans from a standard Gaussian distribution and then generates the named entities by denoising them with the learned reverse diffusion process.

Setup

To run our code, install:

conda create -n diffusionner python=3.8
pip install -r requirements.txt

Quick Start

Datasets

Nested NER:

ACE04: https://catalog.ldc.upenn.edu/LDC2005T09
ACE05: https://catalog.ldc.upenn.edu/LDC2006T06
GENIA: http://www.geniaproject.org/genia-corpuss

Flat NER:

OntoNotes: https://github.com/yhcc/OntoNotes-5.0-NER
CoNLL03: https://data.deepai.org/conll2003.zip
MSRA: https://github.com/ShannonAI/mrc-for-flat-nested-ner/blob/master/ner2mrc/download.md

We provide the preprocessed datasets in these links: ACE2004, GENIA, CoNLL03, MSRA. Please download them and put them into the data/datasets folder.

If you need other datasets, please contact me (syl@zju.edu.cn) by email. Note that you need to state your identity and prove that you have obtained the license.

Training

Take the ACE2004 dataset as a demo and run:

CUBLAS_WORKSPACE_CONFIG=:4096:8 python diffusionner.py train --config configs/ace2004.conf

We also provide the pre-trained checkpoints in these links: ACE2004, GENIA, CoNLL03, MSRA. Please download them and put them into the data/checkpoints folder.

Evaluating

Set the path of the model checkpoint into eval.conf -> model_path and run:

CUBLAS_WORKSPACE_CONFIG=:4096:8 python diffusionner.py eval --config configs/eval.conf

Citation

If you find this repository useful, please cite our paper:

@article{Shen2023DiffusionNERBD,
  title={DiffusionNER: Boundary Diffusion for Named Entity Recognition},
  author={Yongliang Shen and Kaitao Song and Xu Tan and Dong Sheng Li and Weiming Lu and Yue Ting Zhuang},
  journal={ArXiv},
  year={2023},
  volume={abs/2305.13298}
}

Acknowledgement

denoising-diffusion-pytorch

tricktreat/DiffusionNER