CNN and DailyMail News Articles with Summaries

Over 300,000 news articles from CNN and the DailyMail are paired with human-written multi-sentence summaries. The dataset, created by nschantz21 and hosted on Hugging Face, was last updated in July 2022. It is a processed version of the CNN/DailyMail dataset commonly used for abstractive text summarization tasks.

Use Cases

• Train sequence-to-sequence models to generate article summaries from the full text.
• Benchmark extractive summarization algorithms by comparing generated highlights to the reference summaries.
• Fine-tune large language models for news summarization using the article-summary pairs.
• Analyze the structure and style of human-written summaries across different news sources.

Strengths

• Contains over 300,000 unique news articles, a standard scale for model training.
• Includes human-generated abstractive summaries, not just extractive highlights.
• Processed into a structured format (Parquet) for efficient loading with libraries like Polars and Pandas.

Limitations

• Dataset is static with a last update in 2022, lacking recent news events.
• Source articles are primarily from two US/UK news outlets, introducing geographic and editorial bias.
• Specific details on potential class imbalance or label noise in summaries are unavailable.

Provenance

Source

Original articles from CNN and DailyMail websites.

Collection Method

Articles and summaries were scraped and processed from the web, then parsed and formatted for machine learning.

Freshness

Last updated 2022-07-26; static dataset.

Geography

Primarily United States and United Kingdom, based on news outlet origins.