Newspaper Text Summarization CNN DailyMail
Dataset containing over 300,000 news articles in English from CNN and Daily Mail, with their handwritten summaries written by journalists. It supports extractive and abstractive summarization.
Approximately 312,000 articles in English, JSON/text format, including full article and highlights
CC0: Public Domain
Description
The dataset Newspaper Text Summarization CNN DailyMail includes over 300,000 news articles in English from CNN and the Daily Mail. Each article is accompanied by a summary (“highlights”) written by the journalist, allowing the training of automatic summarization models. The dataset is structured in three parts: train, validation and test, to facilitate evaluation.
What is this dataset for?
- Train automatic, extractive, or abstractive summarization models
- Evaluate the quality of summaries generated by NLP systems
- Studying the automatic comprehension of long texts
Can it be enriched or improved?
Yes, by adding additional annotations, for example metadata about journalistic style, or by creating multilingual versions. Cleanups can improve the quality of the input text, and data augmentation techniques can enrich the corpus.
🔎 In summary
🧠 Recommended for
- NLP researchers
- AI developers
- Machine learning students
🔧 Compatible tools
- Hugging Face Transformers
- TensorFlow
- PyTorch
- SpacY
- AllennLP
💡 Tip
For best results, use the train/validation/test splits provided and consider the evaluation with RED.
Frequently Asked Questions
Does this dataset contain articles in multiple languages?
No, only articles in English, mostly from the United States and the United Kingdom.
Can this dataset be used to automatically generate summaries?
Yes, this is the main objective, to train and test extractive and abstractive summarization models.
What is the approximate size of the dataset?
Approximately 312,000 articles with abstracts, totaling over 1 GB of textual data.
