The Curation Corpus is a collection of 40,000 professionally-written summaries of news articles, with links to the articles themselves. This repository provides a scraper to access them. If you're interested in commercial use or access to the wider catalogue of Curation data, including a larger set of over 150,000 professionally-written abstracts and a scalable, on-demand content abstraction API (driven by humans or AI), please get in touch. For our thoughts on how we hope this release will help the NLP community, see our post introducing the dataset.
Documents | License | Avg. summary length (words) | Avg. document length (words) | Avg. summary length (sentences) | Avg. document length (sentences) | Type | |
---|---|---|---|---|---|---|---|
CNN | 90,266 | N/A | 45.7 | 760.5 | 3.59 | 34 | Implied by "summary" box |
DailyMail | 196,961 | N/A | 54.7 | 653.3 | 3.86 | 29.3 | Implied by bullets below headline |
NYT | 110,540 | Non-commerical | 45.5 | 800 | 2.44 | 35.6 | Abstractive summary |
Xsum | 276,711 | N/A | 23.3 | 431 | 1 | 19.7 | Single sentence answering "what is this article about?" |
Curation Base | 40,000 | CC-BY | 82.6 | 527.9 | 4.9 | 27.4 | Professionally written and edited standalone summary intended to be understood by itself |
Curation Large | 134,849* | Commercial | 81.3 | 521 | 4.9 | 27 | Professionally written and edited standalone summary intended to be understood by itself |
- Clone this repository (or just copy the code from scraper.py)
- Download the urls, headlines, and summaries from here
- Run
web_scraper.py
. Give as command line arguments the path to the csv file without article text, the path to a new csv file which will have article text, and a batch size to determine how many urls it will scrape at a time. Larger batch sizes will make it run faster but it may drop more articles due to timeouts. I recommend ~50 on a 2015 Macbook Pro.
git clone https://github.com/CurationCorp/curation-corpus.git
cd curation-corpus
wget https://curation-datasets.s3-eu-west-1.amazonaws.com/curation-corpus-base.csv
python web_scraper.py curation-corpus-base.csv curation-corpus-base-with-articles.csv 50
Some urls will return messy results due to content changing over time, paywalls, etc. We've tried to remove the worst offenders from this release. There is probably still scope though for improving the scraper though.
We are still learning about this field ourselves and will share our tutorials in the examples folder. If you use our dataset in your own research, write a tutorial, or have anything you would like to share, let us know and we will link to it from here!
Curation is a SaaS business combining machine learning & human intelligence enabling executives to effortlessly follow emerging risks, themes and client activity with a particular focus on ESG-related issues. We enable businesses to act faster, delivering significant time and cost savings.
@misc{curationcorpusbase:2020,
title={Curation Corpus Base},
author={Curation Corporation},
year={2020}
}
Please remember to attribute any derivative works in accordance with the terms of the CC-BY license.
This work is licensed under a Creative Commons Attribution 4.0 International License.