AG News: 1 Million Articles for Topic Classification

AG News features over 1 million news articles aggregated from 2,000+ sources by the ComeToMyHead academic search engine. Collected over a one-year period starting in July 2004, it serves as a foundational resource for text classification and information retrieval research.

Use Cases

• Training topic classification models using the 'label' field to distinguish between news categories
• Developing text representation models using the 'text' content
• Benchmarking clustering algorithms on the 'text' field to group similar news stories

Strengths

• 1 million+ news articles
• Sourced from 2,000+ distinct news outlets
• Established academic benchmark for text classification

Limitations

• Temporal staleness with data limited to the 2004-2005 period
• Standard versions often collapse 2,000 sources into only 4 broad labels, reducing granularity

Provenance

Source

ComeToMyHead academic news search engine

Collection Method

Scraped from 2,000+ news sources via an academic search engine

Time Range

2004-2005

Freshness

Static dataset; data collection concluded circa 2005.

Geography

Global (English language)