license: mit
task_categories:
- text-classification
language:
- en
tags:
- CENIA
- News
size_categories:
- 100K<n<1M
This dataset presents a collection of U.S. news headlines between 2019 and 2022 from Twitter and Facebook, with automatic annotations and human verifications. For more details on the methodology and annotation process, please refer to our paper {News Gathering: Leveraging Transformers to Rank News}.
Attributes:
This dataset comprises five attributes: the first corresponds to "Headlines 1," the second to "Headlines 2," the third to the "target" variable {0, 1}, the fourth is related to the "split" {train, validation, test}, and finally, the fifth attribute indicates the "type" {soft label, human-verified}. Both sentences are associated with news extracted from diverse U.S. news sources {The New York Times, San Francisco Chronicle, National Broadcasting, Yahoo News, among other outlets}. The "target" variable indicates whether both sentences relate to the same event {1} or not {0}. Regarding the "type," {soft label} corresponds to an automatic process following the methodology in the paper, and {human-verified} indicates a process verified through a survey by humans.
Data Source:
The primary sources included Twitter, accessed via the Academic API, and Facebook, accessed through CrowdTangle. This facilitated the automatic annotation of USA news articles spanning 2019 to 2022 by the methodology outlined in the paper. Within the test dataset, this sentence pair underwent human verification through a survey.
Data Format:
The dataset is presented in tabular format and comprises five columns, as mentioned in the previous description.