MELALab / nela-gt

Repository for the NELA dataset

NELA-GT repository

This repository contains usage examples for the NELA-GT-2020 data set with Python 3.

NELA-GT-2022

Metadata
Dataset name	nela-gt-2022
Formats	Sqlite3,JSON
No. of articles	1778361
No. of sources	361
No. of embedded tweets	346283
No. of articles w/ tweets	137150
Collection period	2022-01-01 to 2022-12-31

NELA-GT-2021

If you use this dataset in your work, please cite us as follows:

@misc{
    gruppi2020nelagt2021,
    title={NELA-GT-2021: A Large Multi-Labelled News Dataset for The Study of Misinformation in News Articles},
    author={Maurício Gruppi and Benjamin D. Horne and Sibel Adalı},
    year={2021},
    eprint={---},
    archivePrefix={arXiv},
    primaryClass={cs.CY}
}

Data

Metadata
Dataset name	nela-gt-2021
Formats	Sqlite3,JSON
No. of articles	1856509
No. of sources	367
No. of embedded tweets	405449
No. of articles w/ tweets	153663
Collection period	2021-01-01 to 2021-12-31

Download

• News Data
  • Full dataset Sqlite3 | JSON

Limitations

Since the articles collected from news sources may be copyrighted, we apply a transformation to the original text so that it cannot be used for their originally intended purpose, i.e., that of being read by individuals to consume journalistic information.

We modify the text so that it cannot properly be used for news consumption but that can still be used for text analysis via a transformation.

For articles with more than 200 tokens, we replace 7 tokens with @ every 100 tokens. For articles with fewer than 200 tokens, we replace 5 consecutive tokens with @ every 20 tokens. This transforms the articles so that it is unlikely that a user will read NELA-GT to consume news while still keeping most of the content that is useful for analysis (~7% for larger articles).

Tables

Table: Newsdata

Each data point collected corresponds to an article and contains the fields described below.

Field	Type	Description
id	string	ID of the article.
date	string	date of publication (YYYY-MM-DD).
source	string	name of the source.
title	string	article's headline.
content	string	article's body text.
author	string	author who signed the article.
published	string	date time string as provided by source.
published_utc	integer	unix timestamp of publication.
collection_utc	integer	unix timestamp of collection date.
url	string	url of the paper.

Table: Tweet

Each entry corresponds to an embedded tweet observed in the article with id article_id.

Field	Type	Description
id	string	ID of the embedded tweet.
article_id	string	ID of the article that contains the embedded tweet.
embedded_tweet	string	ID/URL of the embedded tweet.

Aggregated labels

We provide aggregated labels based on Media Bias/Fact Check reports, classifying each source as:

• Reliable - class 0
• Unreliable - class 1
• Mixed - class 2
• Null - invalid label, -1 or null

These labels can be found in labels.csv

Note: the labels used in this aggregation were collected from Media Bias/Fact Check on Mar 20, 2020.

Examples

Please refer to these examples for details on how to use our dataset using Python3 and Pandas.

load-sqlite3.py

• How to load the data from the Sqlite3 database using SQL queries.
  • Loading data from single or multiple sources from the database
  • Loading data from the database into a Pandas dataframe

Usage:

python3 load-sqlite3.py <path-to-database>

load-json.py

• How to load NELA in JSON format with Python 3.
  • Loading a single source's JSON
  • Loading a directory of NELA JSON files - WARNING: this consumes a lot of memory

Usage:

python3 load-json.py <path-to-file>

About NELA-GT-2020

Citation

If you use this dataset in your work, please cite us as follows:

@misc{
    gruppi2020nelagt2020,
    title={NELA-GT-2020: A Large Multi-Labelled News Dataset for The Study of Misinformation in News Articles},
    author={Maurício Gruppi and Benjamin D. Horne and Sibel Adalı},
    year={2021},
    eprint={---},
    archivePrefix={arXiv},
    primaryClass={cs.CY}
}

Data

We release our main news dataset NELA-GT-2020 along with two subsets, created by doing keyword searches on the main dataset. We introduce the NELA-GT-ELECTIONS dataset, containing articles related to the 2020 U.S. Presidential Elections, and the NELA-GT-COVID19 subset, which contains articles related to the COVID-19 pandemic.

Metadata
Dataset name	NELA-GT-2020	NELA-GT-ELECTIONS	NELA-GT-COVID19
Formats	Sqlite3,JSON	Sqlite3, JSON	Sqlite3, JSON
No. of articles	1779127	294504	699803
No. of sources	519	403	493
No. of embedded tweets	410784	107771	158855
Collection period	2020-01-01 to 2020-12-31	2020-01-01 to 2020-12-31	2020-01-01 to 2020-12-31

Download

• News Data

  • Full dataset Sqlite3 | JSON
  • COVID-19 subset Sqlite3 | JSON
  • U.S. elections subset Sqlite3 | JSON

• Source Labels: CSV

  • This file contains the credibility label for news sources in the dataset (reliable, unreliable, mixed).

For more details about this dataset, see the paper.