metadata

license: apache-2.0
annotations_creators:
  - expert-generated
language_creators:
  - found
task_categories:
  - text-classification
language:
  - en
multilinguality:
  - monolingual
source_datasets:
  - Opensources https://github.com/BigMcLargeHuge/opensources
  - FakeNews Corpus https://github.com/several27/FakeNewsCorpus
tags:
  - fake-news-detection
  - fake news
  - english
  - nlp
task_ids:
  - topic-classification
  - fact-checking
pretty_name: Fake News Opensources
size_categories:
  - 1M<n<10M
dataset_info:
  features:
    - name: id
      dtype: int
    - name: type
      dtype: string
    - name: domain
      dtype: string
    - name: scraped_at
      dtype: string
    - name: url
      dtype: string
    - name: authors
      dtype: string
    - name: title
      dtype: string
    - name: content
      dtype: string

Dataset Card for "Fake News Opensources"

Dataset Description
Dataset Structure
Dataset Creation
Considerations for Using the Data
Additional Information

Dataset Description

Homepage: https://github.com/AndyTheFactory/FakeNewsDataset
Repository: https://github.com/AndyTheFactory/FakeNewsDataset
Point of Contact: Andrei Paraschiv

Dataset Summary

a consolidated and cleaned up version of the opensources Fake News dataset Fake News Corpus (cite) comprises 8,529,090 individual articles, classified into 12 classes: reliable, unreliable, political, bias, fake, conspiracy, rumor clickbait, junk science, satire, hate and unknown. The articles were scraped between the end of 2017 and the beginning of 2018 from various news websites, totaling 647 distinct sources, collecting articles dating from various years leading to the 2016 US elections and the year after. Documents were classified based on their source, based on the curated website list provided by opensources.co using a leading to a high imbalanced class distribution.

This consolidated and cleaned Fake News Corpus Dataset is a refined and enhanced version of the "opensources.co Fake News dataset". This dataset, is a valuable resource for research and analysis in the domain of misinformation, disinformation, and fake news detection.

The oroginal Fake News Corpus comprises a vast collection of 8,529,090 individual articles, categorized into 12 distinct classes: reliable, unreliable, political, bias, fake, conspiracy, rumor clickbait, junk science, satire, hate and unknown.

The articles within this dataset were systematically gathered by the original authors through web scraping activities conducted between the end of 2017 and the beginning of 2018. These articles were sourced from a diverse range of news websites, resulting in a compilation from 647 distinct online sources.

Documents were classified based on their source, using a curated website list provided by opensources.co, a crowdsourced platform for tracking online information sources. Their proposed source classification method, was based on six criteria:

Title and Domain name analysis,
“About Us” analysis,
source or study mentioning,
writing style analysis,
aesthetic analysis and social media analysis.

After extensive data cleaning and duplicate removal we retain 5,915,569 records

Languages

English

Dataset Structure

Data Instances

An example record looks as follows.

{
  'id': 4059480,
  'type': 'political',
  'domain': 'dailycaller.com',
  'scraped_at': '2017-11-27',
  'url': 'http://dailycaller.com/buzz/massachusettsunited-states/page/2/',
  'authors': 'Jeff Winkler, Jonathan Strong, Ken Blackwell, Pat Mcmahon, Julia Mcclatchy, Admin, Matt Purple',
  'title': 'The Daily Caller',
  'content':'New Hampshire is the state with the highest median income in the nation, according to the U.S. Census Bureauâ€™s report on income, poverty and health insurance',
}

Data Fields

id: The unique article ID
type: the label of the record (one of: reliable, unreliable, political, bias, fake, conspiracy, rumor clickbait, junk science, satire, hate)
'scraped_at': date of the original scrape run
'url': original article url
'authors': comma separated list of scraped authors
'title': original scraped article title
content: full article text

Data Splits

Label	Nr Records
reliable	1807323
political	968205
bias	769874
fake	762178
conspiracy	494184
rumor	375963
unknown	230532
clickbait	174176
unreliable	104537
satire	84735
junksci	79099
hate	64763

total	5915569

Dataset Creation

Curation Rationale

Source Data

News Articles comments

Initial Data Collection and Normalization

Who are the source language producers?

Sports News Article readers

Annotations

Annotation process

Who are the annotators?

Journalists

Personal and Sensitive Information

Considerations for Using the Data

Social Impact of Dataset

Discussion of Biases

Other Known Limitations

The dataset was not manually filtered, therefore some of the labels might not be correct and some of the URLs might not point to the actual articles but other pages on the website. However, because the corpus is intended for use in training machine learning algorithms, those problems should not pose a practical issue.

Additionally, when the dataset will be finalised (as for now only about 80% was cleaned and published), I do not intend to update it, therefore it might quickly become outdated for other purposes than content-based algorithms. However, any contributions are welcome!

Additional Information

Dataset Curators

Licensing Information

This data is available and distributed under Apache-2.0 license

Citation Information

tbd

Datasets:

andyP
/

fake_news_en_opensources