Optimizing Bi-Encoder for Named Entity Recognition via Contrastive Learning

Introduction

This is the repository for BINDER (BI-encoder for NameD Entity Recognition via Contrastive Learning) accepted at ICLR 2023.

BINDER employs two encoders to separately map text and entity types into the same vector space, and reuses the vector representations of entity types for different text spans (or vice versa), resulting in a faster training and inference speed. Based on the bi-encoder representations, BINDER introduces a unified contrastive learning framework for NER, which encourages the representation of entity types to be similar with the corresponding entity mentions, and to be dissimilar with non-entity text spans. BINDER also introudces a novel dynamic thresholding loss in contrastive learning. At test time, it leverages candidate-specific dynamic thresholds to distinguish entity spans from non-entity ones. Check out our paper for the details.

If you find our code is useful, please cite:

@article{zhang-etal-2022-binder,
  title={Optimizing Bi-Encoder for Named Entity Recognition via Contrastive Learning},
  author={Zhang, Sheng and Cheng, Hao and Gao, Jianfeng and Poon, Hoifung},
  journal={arXiv preprint arXiv:2208.14565},
  year={2022}
}

Quick Start

1. Data Preparation

Follow the instructions README.md in the data_preproc folder.

2. Environment Setup

conda create -n binder -y python=3.9
conda activate binder
conda install pytorch==1.13 pytorch-cuda=11.6 -c pytorch -c nvidia
pip install transformers==4.24.0 datasets==2.6.1 wandb==0.13.5 seqeval==1.2.2

3. Experiment Run

Assuming you have prepared data for ACE2005 and finished environment setup, below is the command to run an experiment on ACE2005:

python run_ner.py conf/ace05.json

To run experiments on other datasets, simply change the config.

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.

microsoft / binder

readme