AskDocs: A medical QA dataset

ELI5 dataset adapted on Medical Questions (AskDocs) subreddit.

Getting Started

	Train	Valid	Test	External
en	24256	5198	5198	166804
pt	24256	5198	5198	166804

The dataset questions and answers span a period from January 2013 to December 2019.

We additionally translated to Portuguese and used external data from here, which is a binary classification dataset "a QNLI medical-like". We adapted to value 5 or 0.

Usage

Datasets 🤗

from datasets import load_dataset

data = load_dataset("ju-resplande/askD", split="train_pt")
# ['train_en', 'validation_en', 'test_en', 'external_en', 'train_pt', 'validation_pt', 'test_pt', 'external_pt']

Citing

@misc{Gomes20202,
  author = {GOMES, J. R. S.},
  title = {AskDocs: A medical QA dataset},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/ju-resplande/askD}},
  commit = {42060c4402c460e174cbb75a868b429c554ba2b7}
}

Acknowledgments

@viniciusplo and @ruanchaves for giving the idea. 😃