pretty_name: CC-News
annotations_creators:
- no-annotation
language_creators:
- found
language:
- en
license:
- unknown
multilinguality:
- monolingual
size_categories:
- 100K<n<1M
source_datasets:
- original
task_categories:
- text-generation
- fill-mask
task_ids:
- language-modeling
- masked-language-modeling
paperswithcode_id: cc-news
dataset_info:
features:
- name: title
dtype: string
- name: text
dtype: string
- name: domain
dtype: string
- name: date
dtype: string
- name: description
dtype: string
- name: url
dtype: string
- name: image_url
dtype: string
config_name: plain_text
splits:
- name: train
num_bytes: 2016418133
num_examples: 708241
download_size: 845131146
dataset_size: 2016418133
Dataset Card for CC-News
Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- Homepage: CC-News homepage
- Point of Contact: Vladimir Blagojevic
Dataset Summary
CC-News dataset contains news articles from news sites all over the world. The data is available on AWS S3 in the Common Crawl bucket at /crawl-data/CC-NEWS/.
This version of the dataset has been prepared using news-please - an integrated web crawler and information extractor for news.
It contains 708241 English language news articles published between Jan 2017 and December 2019.
It represents a small portion of the English language subset of the CC-News dataset.
Supported Tasks and Leaderboards
CC-News has been mostly used for language model training.
Languages
The text in the dataset is in the English language.
Dataset Structure
Data Instances
Dataset instance contains an article itself and the relevant article fields. An example from the Cc-New train set looks as follows:
{
'date': '2017-08-14 00:00:00',
'description': '"The spirit of Green Day has always been about rising above oppression."',
'domain': '1041jackfm.cbslocal.com',
'image_url': 'https://cbs1041jackfm.files.wordpress.com/2017/08/billie-joe-armstrong-theo-wargo-getty-images.jpg?w=946',
'text': 'By Abby Hassler\nGreen Day’s Billie Joe Armstrong has always been outspoken about his political beliefs. Following
the tragedy in Charlottesville, Virgina, over the weekend, Armstrong felt the need to speak out against the white supremacists
who caused much of the violence.\nRelated: Billie Joe Armstrong Wins #TBT with Childhood Studio Photo\n“My heart feels heavy.
I feel like what happened in Charlottesville goes beyond the point of anger,” Armstrong wrote on Facebook. “It makes me sad
and desperate. shocked. I f—— hate racism more than anything.”\n“The spirit of Green Day has always been about rising above
oppression. and sticking up for what you believe in and singing it at the top of your lungs,” Armstrong continued.
“We grew up fearing nuclear holocaust because of the cold war. those days are feeling way too relevant these days.
these issues are our ugly past.. and now it’s coming to haunt us. always resist these doomsday politicians. and in the
words of our punk forefathers .. Nazi punks f— off.”',
'title': 'Green Day’s Billie Joe Armstrong Rails Against White Nationalists',
'url': 'http://1041jackfm.cbslocal.com/2017/08/14/billie-joe-armstrong-white-nationalists/'
}
Data Fields
date: date of publicationdescription: description or a summary of the articledomain: source domain of the article (i.e. www.nytimes.com)image_url: URL of the article's imagetext: the actual article text in raw formtitle: title of the articleurl: article URL, the original URL where it was scraped.
Data Splits
CC-News dataset has only the training set, i.e. it has to be loaded with train split specified:
cc_news = load_dataset('cc_news', split="train")
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data
Initial Data Collection and Normalization
CC-News dataset has been proposed, created, and maintained by Sebastian Nagel.
The data is publicly available on AWS S3 Common Crawl bucket at /crawl-data/CC-NEWS/.
This version of the dataset has been prepared using news-please - an
integrated web crawler and information extractor for news.
It contains 708241 English language news articles published between Jan 2017 and December 2019.
Although news-please tags each news article with an appropriate language tag, these tags are somewhat unreliable.
To strictly isolate English language articles an additional check has been performed using
Spacy langdetect pipeline.
We selected articles with text fields scores of 80% probability or more of being English.
There are no strict guarantees that each article has all the relevant fields. For example, 527595
articles have a valid description field. All articles have what appears to be a valid image URL,
but they have not been verified.
Who are the source language producers?
The news websites throughout the World.
Annotations
Annotation process
[N/A]
Who are the annotators?
[N/A]
Personal and Sensitive Information
As one can imagine, data contains contemporary public figures or individuals who appeared in the news.
Considerations for Using the Data
Social Impact of Dataset
The purpose of this dataset is to help language model researchers develop better language models.
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
[More Information Needed]
Licensing Information
[More Information Needed]
Citation Information
@InProceedings{Hamborg2017,
author = {Hamborg, Felix and Meuschke, Norman and Breitinger, Corinna and Gipp, Bela},
title = {news-please: A Generic News Crawler and Extractor},
year = {2017},
booktitle = {Proceedings of the 15th International Symposium of Information Science},
location = {Berlin},
doi = {10.5281/zenodo.4120316},
pages = {218--223},
month = {March}
}
Contributions
Thanks to @vblagoje for adding this dataset.