metadata

license: cc-by-nc-sa-4.0
pretty_name: Babel Briefings
language:
  - en
  - es
  - de
  - fr
  - zh
  - ar
  - pt
  - bg
  - cs
  - el
  - he
  - hu
  - id
  - it
  - ja
  - ko
  - lt
  - lv
  - nl
  - 'no'
  - pl
  - ro
  - ru
  - sk
  - sl
  - sr
  - sv
  - th
  - tr
  - uk
task_categories:
  - text-classification
  - translation
  - zero-shot-classification
  - feature-extraction
  - text-generation
tags:
  - news
  - headlines
  - business
  - science
  - technology
  - sports
  - health
  - entertainment
size_categories:
  - 1M<n<10M

Babel Briefings News Headlines Dataset README

Break Free from the Language Barrier

Version: 1 - Date: 30 Oct 2023

Collected and Prepared by Felix Leeb (Max Planck Institute for Intelligent Systems, Tübingen, Germany)

Check out our paper on arxiv.

This dataset contains 4,719,199 news headlines across 30 different languages collected between 8 August 2020 and 29 November 2021. The headlines were collected using the News API by collecting the top headlines (usually about 30-70 articles) separately for each combination of the 54 locations x 7 categories almost every day. Note, that the same article may occur more than once across different locations, categories, or dates (which is recorded in the instances property), so in total 7,419,089 instances were collected.

For non-English articles, the article data is translated to English using Google Translate (see en-title, en-description, and en-content properties).

The dataset is provided in the form of 54 JSON files, one for each location containing the all the unique headlines that appeared for the first time in the corresponding location. Each headline is represented as a JSON object with the following properties:

ID: (integer) a unique ID for each article
title: (string) the headline text in the original language
description: (string) the article description in the original language
content: (string) the first few words of the article in the original language
author: (string) the author of the article
source-id: (string) the news aggregator (e.g. Google-News)
source-name: (string) usually the domain of the source where the article was published
url: (string) the URL of the article
urlToImage: (string) the URL to an image associated with the article
publishedAt: (date) the article was published
instances: (list) specific time and place where this article was posted. Each element contains:
- collectedAt: (date) date and time when the article was collected
- category: (string) of the article from 7 possible values (see below for full list)
- location: (string) of the article from 54 possible values (see below for full list)
language: (string) ISO-639 2-letter code for the language (inferred from location)
en-title: (string) the headline text translated to English (if necessary)
en-description: (string) the article description text translated to English (if necessary)
en-content: (string) the first few words of the article translated to English (if necessary)

Notes

Unfortunately, due to an issue with News API, the content of articles originally in a non-latin based script (e.g. Chinese, Arabic, Japanese, Greek, Russian, etc.) are usually not available. However, for the most part all other articles should have a meaningful content property, and the title and descriptions appear unaffected.
All properties except language, en-title, en-description, and en-content are taken directly from the News API responses. The language is inferred from the location, and the English translations are collected using Google Translate.

Statistics

Here are a few basic summary statistics about the dataset.

Articles by Language

Code	Language	Articles	Locations
en	English	1128233	au, ca, gb, ie, in, my, ng, nz, ph, sa, sg, us, za
es	Spanish	455952	ar, co, cu, mx, ve
fr	French	288328	be, fr, ma
zh	Chinese	270887	cn, hk, tw
de	German	259718	at, ch, de
pt	Portuguese	243829	br, pt
ar	Arabic	178854	ae, eg
id	Indonesian	131252	id
it	Italian	129005	it
tr	Turkish	122724	tr
el	Greek	119940	gr
ja	Japanese	118475	jp
pl	Polish	116904	pl
ru	Russian	113395	ru
nl	Dutch	104031	nl
th	Thai	90708	th
sv	Swedish	86838	se
ko	Korean	83090	kr
sr	Serbian	80040	rs
hu	Hungarian	73509	hu
cs	Czech	70647	cz
he	Hebrew	67794	il
bg	Bulgarian	67223	bg
uk	Ukrainian	65610	ua
ro	Romanian	54601	ro
no	Norwegian	46804	no
sk	Slovak	43057	sk
lv	Latvian	40006	lv
lt	Lithuanian	34719	lt
sl	Slovenian	33026	si

Instances by category

Category	Instances
sports	1132542
entertainment	982479
business	840748
technology	802933
general	704692
health	424188
science	388281

Instances by location

Code	Location	Instances
ae	United Arab Emirates	214256
ar	Argentina	159139
ph	Philippines	155365
ng	Nigeria	155112
in	India	145536
us	United States	144800
ca	Canada	143928
sa	Saudi Arabia	143382
cu	Cuba	138675
au	Australia	138408
br	Brazil	136101
ma	Morocco	131974
id	Indonesia	131252
eg	Egypt	129382
it	Italy	129005
gb	United Kingdom	127391
ie	Ireland	126640
mx	Mexico	124499
tr	Turkey	122724
gr	Greece	119940
de	Germany	119917
jp	Japan	118475
za	South Africa	117351
fr	France	117210
pl	Poland	116904
pt	Portugal	115976
co	Colombia	115325
my	Malaysia	115223
ru	Russian Federation	113395
at	Austria	111867
nz	New Zealand	108809
tw	Taiwan	108652
nl	Netherlands	104031
sg	Singapore	101251
be	Belgium	99460
cn	China	91561
ve	Venezuela	91045
th	Thailand	90708
se	Sweden	86838
kr	Korea	83090
hk	Hong Kong	83051
rs	Serbia	80040
hu	Hungary	73509
cz	Czechia	70647
ch	Switzerland	68846
il	Israel	67794
bg	Bulgaria	67223
ua	Ukraine	65610
ro	Romania	54601
no	Norway	46804
sk	Slovakia	43057
lv	Latvia	40006
lt	Lithuania	34719
si	Slovenia	33026