BAND: Biomedical Alert News Dataset

Zihao Fu,¹ Meiru Zhang,¹ Zaiqiao Meng,^2,1 Yannan Shen,³ David Buckeridge,³ Nigel Collier¹

¹Language Technology Lab, University of Cambridge
²School of Computing Science, University of Glasgow
³School of Population and Global Health, McGill University

About

The Biomedical Alert News Dataset (BAND) is a well-annotated dataset aimed at improving disease surveillance and understanding of disease spread. It includes 1,508 samples from reported news articles, open emails, and alerts, along with 30 epidemiology-related questions. BAND is designed to challenge and advance NLP models in tasks like Named Entity Recognition (NER), Question Answering (QA), and Event Extraction (EE), with a focus on epidemiological analysis.

Dataset

The BAND dataset can be fond under corresponding folders.

Usage

Question Answering

Run Decoder-Only Model

./scripts/train_lm.sh lmrand base

Run Encoder-Decoder Model

./scripts/train_seq2seq.sh band_rand ptm=t5

Event Extraction

Run Encoder-Decoder Model

./scripts/train_outbreak.sh

Run Decoder-only Model

python train_lm.py --name "gpt2"

Named Entity Recognition

Run Token-Based NER Model

./scripts/run_token_ner.sh

Run CRF-Based NER Model

./scripts/run_crf_ner.sh

Run Span-Based NER Model

./scripts/run_span_ner.sh

Citation

If you find our dataset or paper useful, please cite our work:

@inproceedings{band2024,
  title={BAND: Biomedical Alert News Dataset},
  author={Fu, Zihao and Zhang, Meiru and Meng, Zaiqiao and Shen, Yannan and Buckeridge, David and Collier, Nigel},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2024}
}

License

This project is licensed under the MIT License - refer to the LICENSE file for details.