Free News Datasets

Open news data for machine learning, NLP, and sentiment analysis — provided free by Newsdata.io.

A curated collection of free, downloadable news datasets covering business, sports, entertainment, health, science and technology, world politics, COVID-19, the Ukraine conflict, and more — aggregated from a wide range of reliable publishers across many languages and countries. Each archive ships in both CSV and JSON, with rich metadata and AI-extracted annotations (ai_tag, ai_region, ai_org, sentiment) so you can skip the preprocessing and get straight to your model.

Use them as ready-made news corpora for:

• Machine learning — text classification, clustering, embeddings, LLM fine-tuning
• Natural language processing (NLP) — named entity recognition, topic modelling, summarization
• Sentiment analysis — labelled sentiment + per-sentiment confidence stats out of the box
• Trend analysis — surface what publishers are covering across time
• News aggregation and search demos
• Geopolitical and regional analytics — ai_region / ai_org enrichments included
• Journalism and academic research

Available datasets

Every archive expands to a folder containing the same dataset in CSV and JSON form.

Category	File	Size (compressed)
Business news	Business_News.zip	~17 MB
Sports news	Sports_News.7z	~19 MB
Entertainment news	Entertainment_News.zip	~17 MB
Science and technology news	Science_Technology_News.zip	~13 MB
Health news	Health_News.zip	~4 MB
World politics news	World_Politics_News.zip	~6 MB
COVID-19 news	Covid_News.zip	~9 MB
COVID and vaccine news	Covid_and_Vaccine_News.zip	~10 MB
Ukraine news	Ukraine_News.zip	~5 MB

Want a quick look at the schema before downloading? Tiny samples for the news and crypto-news streams live in sample/ (CSV + JSON + XLSX, under 500 KB each).

File formats and schema

Datasets ship as CSV and JSON (samples also include XLSX). Every article shares the same flat schema:

Field	Type	Description
article_id	string	Stable unique identifier for the article
title	string	Article headline
link	string	URL to the original article
description	string	Short summary / lede
content	string	Full article text
keywords	string	Comma-separated keywords
creator	string	Author / byline
pubDate	datetime	Publication date (UTC)
image_url	string	Hero / lead image URL
video_url	string	Embedded video URL, when present
source_id	string	Publisher identifier
source_priority	int	Publisher priority rank
source_url	string	Publisher domain
source_icon	string	Publisher favicon URL
language	string	ISO 639-1 language code (en, es, …)
country	string	ISO 3166-1 alpha-2 country code (us, gb, …)
category	string	Top-level category
ai_tag	string	AI-extracted topic tag(s)
ai_region	string	AI-extracted geographic region
ai_org	string	AI-extracted organisation mention
sentiment	string	positive / neutral / negative
sentiment_stats	object	Per-sentiment confidence scores

Quick start

Clone the whole repository (or download a single archive from GitHub's file view) and unpack the category you need:

git clone https://github.com/newsdataapi/newsdata.io-free-datasets.git
cd newsdata.io-free-datasets

unzip Business_News.zip          # for .zip archives
7z x Sports_News.7z              # for .7z (install p7zip if needed)

Load with Python (pandas)

import pandas as pd

df = pd.read_csv("Business_News/Business_News.csv")
print(df.shape)
print(df[["title", "language", "country", "sentiment"]].head())

Load with Hugging Face Datasets

from datasets import load_dataset

ds = load_dataset("csv", data_files="Business_News/Business_News.csv")
print(ds["train"].features)

Load with Node.js

import fs from 'node:fs';

const data = JSON.parse(fs.readFileSync('Business_News/Business_News.json'));
console.log(data.length, 'articles');

Why these datasets?

• Free and openly licensed — released under CC BY 4.0, usable in commercial and non-commercial projects with simple attribution
• Multi-category — nine distinct verticals from finance to sport to geopolitics
• Multi-language and multi-country — sources from publishers around the world
• AI-enriched out of the box — every article ships with AI-generated tags, region/organisation mentions, and sentiment scores, so you can skip a lot of preprocessing
• Two formats included — pick CSV for tabular tools (pandas, R, BI), JSON for streaming and NoSQL workflows; samples also include XLSX
• Schema-matched to a live API — when you need fresh data, the same schema is available in real time via the Newsdata.io REST API

Need real-time or custom news data?

These archives are static snapshots. For real-time and historical news in the same schema, use the Newsdata.io API or one of the official SDKs:

• REST API documentation — https://newsdata.io/documentation
• Python SDK — newsdataapi on PyPI
• PHP SDK — newsdataio/newsdataapi on Packagist
• Node.js SDK — newsdata-nodejs-client on npm
• React SDK — newsdataapi on npm

Sign up for a free Newsdata.io API key to get started — no credit card required.

About Newsdata.io

Newsdata.io provides high-quality real-time and historical news data through a developer-friendly REST API, sourced from thousands of reliable publishers across 200+ countries and 75+ languages. Our mission is to make news data easily accessible for analytics, research, journalism, and AI/ML applications.

License

These datasets are released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license — see LICENSE for the full text.

You are free to share (copy and redistribute) and adapt (remix, transform, build upon) the datasets for any purpose, including commercial use, provided you give appropriate attribution to Newsdata.io and indicate if changes were made.

A simple attribution line is enough, e.g.:

News data provided by Newsdata.io, licensed under CC BY 4.0.

Note on the underlying news content: The articles aggregated in these datasets were originally authored by their respective publishers and remain subject to those publishers' own copyrights. The CC BY 4.0 license here applies to the dataset compilation (a "sui generis" database right under CC BY 4.0 §4) — please respect the original publishers' rights when redistributing or republishing individual articles.

---

Found these datasets useful? Star the repo to help other researchers and developers discover it. Feedback and dataset requests are welcome via Issues.