Datasets:
YAML tags: null
annotations_creators:
- expert-generated
language_creators: []
language:
- en
license: []
multilinguality:
- monolingual
pretty_name: DanL/scientific-challenges-and-directions-dataset
source_datasets:
- CORD-19
task_categories:
- text-classification
task_ids:
- multi-label-classification
Dataset Card for scientific-challenges-and-directions
Table of Contents
- Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- Repository: repo
- Paper: A Search Engine for Discovery of Scientific Challenges and Directions
- Point of Contact: lahav@mail.tau.ac.il,tomh@allenai.org
Dataset Summary
The scientific challenges and directions dataset is a collection of 2894 sentences and their surrounding contexts, from 1786 full-text papers in the CORD-19 corpus, labeled for classification of challenges and directions by expert annotators with biomedical and bioNLP backgrounds.
At a high level, our labels are defined as follows:
- Challenge: A sentence mentioning a problem, difficulty, flaw, limitation, failure, lack of clarity, or knowledge gap.
- Research direction: A sentence mentioning suggestions or needs for further research, hypotheses, speculations, indications or hints that an issue is worthy of exploration.
The dataset was developed to help scientists and medical professionals discover challenges and potential directions across scientific literature.
Languages
The language in the dataset is English as written by authors of the scientific papers in the CORD-19 corpus.
Dataset Structure
Data Instances
For each instance, there is a unique id, a string for the text sentence, a string for the previous sentence, a string for the next sentence, and a list for the challenge and direction labels.
{'id': 'PMC7152165_152',
'label': [0.0, 0.0],
'next_sent': 'The railways brought a new technology and vast engineering and architectural structures into Britain’s rural and urban landscapes.',
'prev_sent': 'In Britain, improvements in coaching technologies and roads helped to increase stage coach speeds in the late eighteenth and early nineteenth centuries, while the railway construction boom of the 1830s and 1840s led to a massive reduction in journey times, and the emergence of distinctly new experiences and geographies.',
'text': 'Britain’s railway companies were among the nation’s largest employers in the nineteenth century, and they facilitated the mobility of passengers and important commodities.'}
Data Fields
- id: A string as a unique id for the instance. The id is composed of the unique PMC id of the paper, an underscore, and the index of the sentence within the paper.
- next_sent_: A string of a sentence that is following the text of the instance. If the text is the first in its paragraph the string is saved as '|'.
- prev_sent_: A string of a sentence that is preceding the text of the instance. If the text is the first in its paragraph the string is saved as '|'.
- text: A string of the sentence we seek to classify.
- label: A list of 2 values - the first is the label for challenge and the last of direction. Each value may be either 0, indicating that the text is not challenge or direction, or 1, indicating that the the text is challenge or direction. Each instance can be a challenge, a direction, both, or neither.
Data Splits
The scientific-challenges-and-directions dataset has 3 splits: train, dev, and test. Each instances shows up in only one split. The splits are stratified with no overlap in papers.
| Labels | Train | Dev | Test | All |
|---|---|---|---|---|
| Not Challenge, Not Direction | 602 | 146 | 745 | 1493 |
| Not Challenge, Direction | 106 | 25 | 122 | 253 |
| Challenge, Not Direction | 288 | 73 | 382 | 743 |
| Challenge, Direction | 155 | 40 | 210 | 405 |
Dataset Creation
Curation Rationale
The resource was developed to help scientists and medical professionals discover challenges and potential directions across scientific literature, focusing on a broad corpus pertaining to the COVID-19 pandemic and related historical research.
Source Data
Initial Data Collection and Normalization
See section 3.1 in our paper.
Who are the source language producers?
The authors of the subset of full-text papers in the CORD-19 dataset, which at the time of creating our dataset included roughly 180K documents.
Annotations
Annotation process
See section 3.1 in our paper.
Who are the annotators?
Four expert annotators with biomedical and bioNLP backgrounds. For more details see section 3.1 in our paper.
Personal and Sensitive Information
The dataset does not contain any personal information about the authors or annotators.
Considerations for Using the Data
Social Impact of Dataset
As mentioned, the dataset was developed to help scientists and medical professionals discover challenges and potential directions across scientific literature, focusing on a broad corpus pertaining to the COVID-19 pandemic and related historical research. Studies were conducted to evaluate the utility of the dataset for researchers and medical professionals, in which a prototype based on the dataset was found to outperform other biomedical search tools. For more details see section 4 in our paper. This dataset was also developed for evaluating representational systems for scientific text classification and can be used as such.
Discussion of Biases
The source of the dataset is the full-text papers in the CORD-19 dataset, so biases in CORD-19 may be replicated to our dataset.
Other Known Limitations
N/A
Additional Information
Dataset Curators
The dataset was developed by Dan Lahav, Jon Saad Falcon, Bailey Kuehl, Sophie Johnson, Sravanthi Parasa, Noam Shomron, Duen Horng Chau, Diyi Yang, Eric Horvitz, Daniel S. Weld and Tom Hope as part of Tel Aviv University, the Allen Institute for AI, University of Washington, Georgia Institute of Technology, Microsoft and Swedish Medical Group.
It was supported by the Edmond J. Safra Center for Bioinformatics at Tel-Aviv University, ONR grant N00014-18-1-2193, NSF RAPID grant 2040196, the WR-F/Cable Professorship, and AI2.
Licensing Information
[More Information Needed]
Citation Information
If using our dataset and models, please cite:
@misc{lahav2021search,
title={A Search Engine for Discovery of Scientific Challenges and Directions},
author={Dan Lahav and Jon Saad Falcon and Bailey Kuehl and Sophie Johnson and Sravanthi Parasa and Noam Shomron and Duen Horng Chau and Diyi Yang and Eric Horvitz and Daniel S. Weld and Tom Hope},
year={2021},
eprint={2108.13751},
archivePrefix={arXiv},
primaryClass={cs.CL}
}