Dataset · shash42
10-year daily news archive
shash42/forecast-news
Cleaned corpus of 31.8 million news articles for 2016-2026 for forecasting systems.
What's inside
The repository contains 31,859,020 deduplicated news articles from August 26, 2016 to August 31, 2026, divided into 3,463 daily sections. The total data volume is about 158.5 GiB; the main storage format is parquet, with daily JSONL files and compact JSON with headlines also available.
What you can do with it
- Training and testing event forecasting models and future simulators
- Analyzing news trends and time series over a decade horizon
- Pretraining or fine-tuning language models on news vocabulary
- Size
- 10M–100M rows
- Downloads this month
- 343.7K
- Likes
- 9
How to load it
from datasets import load_dataset
ds = load_dataset("shash42/forecast-news")