Newspaper4k Project grew from a fork of the well known newspaper3k by codelucas which was not updated since September 2020. The initial goal of this fork was to keep the project alive and to add new features and fix bugs. As of version 0.9.3 there are many new features and improvements that make Newspaper4k a great tool for article scraping and curation. To make the migration to Newspaper4k easier, all the classes and methods from the original project were kept and the new features were added on top of them. All API calls from the original project still work as expected, such that for users familiar with newspaper3k you will feel right at home with Newspaper4k.
At the moment of the fork, in the original project were over 400 open issues, which I have duplicated, and as of v 0.9.3 only about 180 issues still need to be verified (many are already fixed, but it's pretty cumbersome to check - hint hint ... anyone contributing?). If you have any issues or feature requests please open an issue here.
| Experimental ChatGPT helper bot for Newspaper4k: |  |
- Python 3.8+ minimum
pip install newspaper4kYou can start directly from the command line, using the included CLI:
python -m newspaper --url="https://edition.cnn.com/2023/11/17/success/job-seekers-use-ai/index.html" --language=en --output-format=json --output-file=article.json
More information about the CLI can be found in the CLI documentation.
Alternatively, you can use Newspaper4k in Python:
import newspaper
article = newspaper.article('https://edition.cnn.com/2023/10/29/sport/nfl-week-8-how-to-watch-spt-intl/index.html')
print(article.authors)
# ['Hannah Brewitt', 'Minute Read', 'Published', 'Am Edt', 'Sun October']
print(article.publish_date)
# 2023-10-29 09:00:15.717000+00:00
print(article.text)
# New England Patriots head coach Bill Belichick, right, embraces Buffalo Bills head coach Sean McDermott ...
print(article.top_image)
# https://media.cnn.com/api/v1/images/stellar/prod/231015223702-06-nfl-season-gallery-1015.jpg?c=16x9&q=w_800,c_fill
print(article.movies)
# []
article.nlp()
print(article.keywords)
# ['broncos', 'game', 'et', 'wide', 'chiefs', 'mahomes', 'patrick', 'denver', 'nfl', 'stadium', 'week', 'quarterback', 'win', 'history', 'images']
print(article.summary)
# Kevin Sabitus/Getty Images Denver Broncos running back Javonte Williams evades Green Bay Packers safety Darnell Savage, bottom.
# Kathryn Riley/Getty Images Kansas City Chiefs quarterback Patrick Mahomes calls a play during the Chiefs' 19-8 Thursday Night Football win over the Denver Broncos on October 12.
# Paul Sancya/AP New York Jets running back Breece Hall carries the ball during a game against the Denver Broncos.
# The Broncos have not beaten the Chiefs since 2015, and have never beaten Chiefs quarterback Patrick Mahomes.
# Australia: NFL+, ESPN, 7Plus Brazil: NFL+, ESPN Canada: NFL+, CTV, TSN, RDS Germany: NFL+, ProSieben MAXX, DAZN Mexico: NFL+, TUDN, ESPN, Fox Sports, Sky Sports UK: NFL+, Sky Sports, ITV, Channel 5 US: NFL+, CBS, NBC, FOX, ESPN, Amazon Prime
This way you can build a Source object from a newspaper websites. This class will allow you to get all the articles and categories on the website. When you build the source, articles are not yet downloaded. The build() call will parse front page, will detect category links (if possible), get any RSS feeds published by the news site, and will create a list of article links.
You need to call download_articles() to download the articles, but note that it can take a significant time.
download_articles() will download the articles in a multithreaded fashion using ThreadPoolExecutor from the concurrent package. The number of concurrent threads can be configured in Configuration.number_threads or passed as an argument to build().
import newspaper
cnn_paper = newspaper.build('http://cnn.com', number_threads=3)
print(cnn_paper.category_urls())
# ['https://cnn.com', 'https://money.cnn.com', 'https://arabic.cnn.com',
# 'https://cnnespanol.cnn.com', 'http://edition.cnn.com',
# 'https://edition.cnn.com', 'https://us.cnn.com', 'https://www.cnn.com']
article_urls = [article.url for article in cnn_paper.articles]
print(article_urls[:3])
# ['https://arabic.cnn.com/middle-east/article/2023/10/30/number-of-hostages-held-in-gaza-now-up-to-239-idf-spokesperson',
# 'https://arabic.cnn.com/middle-east/video/2023/10/30/v146619-sotu-sullivan-hostage-negotiations',
# 'https://arabic.cnn.com/middle-east/article/2023/10/29/norwegian-pm-israel-gaza']
article = cnn_paper.articles[0]
article.download()
article.parse()
print(article.title)
# المتحدث باسم الجيش الإسرائيلي: عدد الرهائن المحتجزين في غزة يصل إلىOr if you want to get bulk articles from the website (have in mind that this could take a long time and could get your IP blocked by the newssite):
import newspaper
cnn_source = newspaper.build('http://cnn.com', number_threads=3)
print(len(newspaper.article_urls))
articles = source.download_articles()
print(len(articles))
print(articles[0].title)First, make sure you have the google extra installed, since we rely on the Gnews package to get the articles from Google News. You can install it using pip like this:
pip install newspaper4k[gnews]Then you can use the GoogleNews class to get articles from Google News:
from newspaper.google_news import GoogleNewsSource
source = GoogleNewsSource(
country="US",
period="7d",
max_results=10,
)
source.build(top_news=True)
print(source.article_urls())
# ['https://www.cnn.com/2024/03/18/politics/trump-464-million-dollar-bond/index.html', 'https://www.cnn.com/2024/03/18/politics/supreme-court-new-york-nra/index.html', ...
source.download_articles()Newspaper can extract and detect languages seamlessly based on the article meta tags. Additionally, you can specify the language for the website / article. If no language is specified, Newspaper will attempt to auto detect a language from the available meta data. The fallback language is English.
Language detection is crucial for accurate article extraction. If the wrong language is detected or provided, chances are that no article text will be returned. Before parsing, check that your language is supported by our package.
from newspaper import Article
article = Article('https://www.bbc.com/zhongwen/simp/chinese-news-67084358')
article.download()
article.parse()
print(article.title)
# 晶片大战：台湾厂商助攻华为突破美国封锁？
if article.config.use_meta_language:
# If we use the autodetected language, this config attribute will be true
print(article.meta_lang)
else:
print(article.config.language)
# zhCheck out The Docs for full and detailed guides using newspaper.
- Multi-threaded article download framework
- Newspaper website category structure detection
- News url identification
- Google News integration
- Text extraction from html
- Top image extraction from html
- All image extraction from html
- Keyword building from the extracted text
- Automatic article text summarization
- Author extraction from text
- Easy to use Command Line Interface (
python -m newspaper....) - Output in various formats (json, csv, text)
- Works in 80+ languages (English, Chinese, German, Arabic, ...) see LANGUAGES.md for the full list of supported languages.
Using the dataset from ScrapingHub I created an evaluator script that compares the performance of newspaper against it's previous versions. This way we can see how newspaper updates improve or worsen the performance of the library.
| Version | Corpus BLEU Score | Corpus Precision Score | Corpus Recall Score | Corpus F1 Score |
|---|---|---|---|---|
| Newspaper3k 0.2.8 | 0.8660 | 0.9128 | 0.9071 | 0.9100 |
| Newspaper4k 0.9.0 | 0.9212 | 0.8992 | 0.9336 | 0.9161 |
| Newspaper4k 0.9.1 | 0.9224 | 0.8895 | 0.9242 | 0.9065 |
| Newspaper4k 0.9.2 | 0.9426 | 0.9070 | 0.9087 | 0.9078 |
| Newspaper4k 0.9.3 | 0.9531 | 0.9585 | 0.9339 | 0.9460 |
Precision, Recall and F1 are computed using overlap of shingles with n-grams of size 4. The corpus BLEU score is computed using the nltk's bleu_score.
We also use our own, newly created dataset, the Newspaper Article Extraction Benchmark (NAEB) which is a collection of over 400 articles from 200 different news sources to evaluate the performance of the library.
| Version | Corpus BLEU Score | Corpus Precision Score | Corpus Recall Score | Corpus F1 Score |
|---|---|---|---|---|
| Newspaper3k 0.2.8 | 0.8445 | 0.8760 | 0.8556 | 0.8657 |
| Newspaper4k 0.9.0 | 0.8357 | 0.8547 | 0.8909 | 0.8724 |
| Newspaper4k 0.9.1 | 0.8373 | 0.8505 | 0.8867 | 0.8682 |
| Newspaper4k 0.9.2 | 0.8422 | 0.8888 | 0.9240 | 0.9061 |
| Newspaper4k 0.9.3 | 0.8695 | 0.9140 | 0.8921 | 0.9029 |
Following system packages are required:
- Pillow:
libjpeg-devzlib1g-devlibpng12-dev - Lxml:
libxml2-devlibxslt-dev - Python Development version:
python-dev
If you are on Debian / Ubuntu, install using the following:
-
Install
python3andpython3-dev:$ sudo apt-get install python3 python3-dev -
Install
pip3command needed to installnewspaper4kpackage:$ sudo apt-get install python3-pip -
lxml requirements:
$ sudo apt-get install libxml2-dev libxslt-dev -
For PIL to recognize .jpg images:
$ sudo apt-get install libjpeg-dev zlib1g-dev libpng12-dev
NOTE: If you find problem installing libpng12-dev, try installing
libpng-dev.
-
Install the distribution via pip:
$ pip3 install newspaper4k
If you are on OSX, install using the following, you may use both homebrew or macports:
$ brew install libxml2 libxslt
$ brew install libtiff libjpeg webp little-cms2
$ pip3 install newspaper4k
see CONTRIBUTING.md
Authored and maintained by Andrei Paraschiv.
Newspaper was originally developed by Lucas Ou-Yang (codelucas), the original repository can be found here. Newspaper is licensed under the MIT license.
Thanks to Lucas Ou-Yang for creating the original Newspaper3k project and to all contributors of the original project.