FiD-NER

Official Code of "Improving Chinese Named Entity Recognition by Search Engine Augmentation"

Requirements

1. torch=1.10.2+cu113

2. transformers=4.17

3. running with CUDA 11.4 and Python 3.9.7

How to run this code

1. download BERT pretrained weights from huggingface to prev_trained_model directory. We use bert-base chinese

2. modify config parameters in scripts/run_ner_crf_xxx.sh

3. run sh scripts/run_ner_crf_xxx.sh

Dataset description

1. Experiments were conducted on three public datasets (Chinese Resume / People's Daily / Weibo NER) and our self-built unconverntional NER datasets from social media BiliBili.
2. "Unconventional" means named entities contains more polysemous words and grammatical ambiguities. Please contact me if you are interested about this dataset.

Citation

If you find FiD-NER interesting and helps your research, please consider citing our work

@article{mao2022improving,
  title={Improving Chinese Named Entity Recognition by Search Engine Augmentation},
  author={Mao, Qinghua and Li, Jiatong and Meng, Kui},
  journal={arXiv preprint arXiv:2210.12662},
  year={2022}
}