MASH-QA Dataset
MASH-QA is a dataset tailored for Question Answering (QA) tasks in the consumer health domain. It is designed to facilitate information extraction from long and comprehensive documents, focusing on answering generally non-factoid questions. Unlike traditional Machine Reading Comprehension (MRC) datasets that feature short, single-span answers, MASH-QA provides answers that are often multi-sentential and sourced from multiple spans of long context articles.
Features of MASH-QA
- Domain: Consumer Health
- Context Length: Long healthcare articles
- Question Type: Generally non-factoid
- Answer Type: Multi-span, multi-sentence, and excerpted from various parts of the context.
Dataset Details
MASH-QA enables researchers to:
- Explore multi-span extraction techniques for long contexts.
- Develop models capable of handling complex QA tasks with long, non-contiguous answers.
Citation
If you use the MASH-QA dataset in your research, please cite the following paper:
@inproceedings{zhu2020question, title={Question Answering with Long Multiple-Span Answers}, author={Ming Zhu and Aman Ahuja and Da-Cheng Juan and Wei Wei and Chandan K. Reddy}, booktitle={Findings of the Association for Computational Linguistics: EMNLP 2020}, pages={3840--3849}, year={2020}, organization={Association for Computational Linguistics} }
Paper
Ming Zhu, Aman Ahuja, Da-Cheng Juan, Wei Wei, and Chandan K. Reddy. 2020. Question Answering with Long Multiple-Span Answers. In Findings of the Association for Computational Linguistics: EMNLP 2020. Association for Computational Linguistics, Online, 3840–3849.
Dataset Access
The dataset is publicly available on GitHub. To access the dataset, visit the following link:
- GitHub Repository: MASH-QA
Usage
The MASH-QA dataset is ideal for tasks such as:
- Long-form Question Answering
- Multi-span Extraction
- Healthcare Informatics Research
License
Please refer to the GitHub repository for license details.
Contact
For any questions or inquiries related to the dataset, please visit the GitHub Issues page in the repository.