Newspaper: Article scraping & curation¶

Inspired by requests for its simplicity and powered by lxml for its speed.

“Newspaper is an amazing python library for extracting & curating articles.” – tweeted by Kenneth Reitz, Author of requests

“Newspaper delivers Instapaper style article extraction.” – The Changelog

We support 10+ languages and everything is in unicode!

>>> import newspaper
>>> newspaper.languages()

Your available langauges are:
input code      full name

  ar              Arabic
  ru              Russian
  nl              Dutch
  de              German
  en              English
  es              Spanish
  fr              French
  he              Hebrew
  it              Italian
  ko              Korean
  no              Norwegian
  pt              Portuguese
  sv              Swedish
  hu              Hungarian
  fi              Finnish
  da              Danish
  zh              Chinese
  id              Indonesian
  vi              Vietnamese

A Glance:¶

>>> import newspaper

>>> cnn_paper = newspaper.build('http://cnn.com')

>>> for article in cnn_paper.articles:
>>>     print(article.url)
u'http://www.cnn.com/2013/11/27/justice/tucson-arizona-captive-girls/'
u'http://www.cnn.com/2013/12/11/us/texas-teen-dwi-wreck/index.html'
...

>>> for category in cnn_paper.category_urls():
>>>     print(category)

u'http://lifestyle.cnn.com'
u'http://cnn.com/world'
u'http://tech.cnn.com'
...

>>> article = cnn_paper.articles[0]

>>> article.download()

>>> article.html
u'<!DOCTYPE HTML><html itemscope itemtype="http://...'

>>> article.parse()

>>> article.authors
[u'Leigh Ann Caldwell', 'John Honway']

>>> article.text
u'Washington (CNN) -- Not everyone subscribes to a New Year's resolution...'

>>> article.top_image
u'http://someCDN.com/blah/blah/blah/file.png'

>>> article.movies
[u'http://youtube.com/path/to/link.com', ...]

>>> article.nlp()

>>> article.keywords
['New Years', 'resolution', ...]

>>> article.summary
u'The study shows that 93% of people ...'

Newspaper has seamless language extraction and detection. If no language is specified, Newspaper will attempt to auto detect a language.

>>> from newspaper import Article
>>> url = 'http://www.bbc.co.uk/zhongwen/simp/chinese_news/2012/12/121210_hongkong_politics.shtml'

>>> a = Article(url, language='zh') # Chinese

>>> a.download()
>>> a.parse()

>>> print(a.text[:150])
香港行政长官梁振英在各方压力下就其大宅的违章建
筑（僭建）问题到立法会接受质询，并向香港民众道歉。
梁振英在星期二（12月10日）的答问大会开始之际在其
演说中道歉，但强调他在违章建筑问题上没有隐瞒的意
图和动机。 一些亲北京阵营议员欢迎梁振英道歉，
且认为应能获得香港民众接受，但这些议员也质问梁振英有

>>> print(a.title)
港特首梁振英就住宅违建事件道歉

If you are certain that an entire news source is in one language, go ahead and use the same api :)

>>> import newspaper
>>> sina_paper = newspaper.build('http://www.sina.com.cn/', language='zh')

>>> for category in sina_paper.category_urls():
>>>     print(category)
u'http://health.sina.com.cn'
u'http://eladies.sina.com.cn'
u'http://english.sina.com'
...

>>> article = sina_paper.articles[0]
>>> article.download()
>>> article.parse()

>>> print(article.text)
新浪武汉汽车综合 随着汽车市场的日趋成熟，传统的“集
全家之力抱得爱车归”的全额购车模式已然过时，另一种轻
松的新兴 车模式――金融购车正逐步成为时下消费者购买
爱车最为时尚的消费理 念，他们认为，这种新颖的购车模
式既能在短期内
...

>>> print(article.title)
两年双免0手续0利率 科鲁兹掀背金融轻松购_武汉车市_武汉
汽车网_新浪汽车_新浪网

Features¶

Works in 10+ languages (English, Chinese, German, Arabic, ...)
Multi-threaded article download framework
News url identification
Text extraction from html
Top image extraction from html
All image extraction from html
Keyword extraction from text
Summary extraction from text
Author extraction from text
Google trending terms extraction

User Guide¶

Installation¶

This part of the documentation covers the installation of newspaper. The first step to using any software package is getting it properly installed.

Distribute & Pip¶

Installing newspaper is simple with pip. However, you will run into fixable issues if you are trying to install on ubuntu.

If you are on Debian / Ubuntu, install using the following:

Python development version, needed for Python.h:
```
$ sudo apt-get install python-dev
```

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

Install the distribution via pip:
```
$ pip install newspaper
```

Download NLP related corpora:

$ curl https://raw.githubusercontent.com/codelucas/newspaper/master/download_corpora.py | python2.7

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

$ pip install newspaper

$ curl https://raw.githubusercontent.com/codelucas/newspaper/master/download_corpora.py | python2.7

Otherwise, install with the following:

NOTE: You will still most likely need to install the following libraries via your package manager

PIL: libjpeg-dev zlib1g-dev libpng12-dev
lxml: libxml2-dev libxslt-dev
Python Development version: python-dev

Note that the Python3 package name is newspaper3k while our Python2 package name is newspaper.

$ pip install newspaper3k

$ curl https://raw.githubusercontent.com/codelucas/newspaper/master/download_corpora.py | python2.7

Get the Code¶

Newspaper is actively developed on GitHub, where the code is always available.

You can clone the public repository:

git clone git://github.com/codelucas/newspaper.git

Once you have a copy of the source, you can embed it in your Python package, or install it into your site-packages easily:

$ pip install -r requirements.txt
$ python setup.py install

Feel free to give our testing suite a shot:

$ python tests/unit_tests.py

Quickstart¶

Eager to get started? This page gives a good introduction in how to get started with newspaper. This assumes you already have newspaper installed. If you do not, head over to the Installation section.

Building a news source¶

Source objects are an abstraction of online news media websites like CNN or ESPN. You can initialize them in two different ways.

Building a Source will extract its categories, feeds, articles, brand, and description for you.

You may also provide configuration parameters like language, browser_user_agent, and etc seamlessly. Navigate to the advanced section for details.

>>> import newspaper
>>> cnn_paper = newspaper.build('http://cnn.com')

>>> sina_paper = newspaper.build('http://www.lemonde.fr/', language='fr')

However, if needed, you may also play with the lower level Source object as described in the advanced section.

Extracting articles¶

Every news source has a set of recent articles.

The following examples assume that a news source has been initialized and built.

>>> for article in cnn_paper.articles:
>>>     print(article.url)

u'http://www.cnn.com/2013/11/27/justice/tucson-arizona-captive-girls/'
u'http://www.cnn.com/2013/12/11/us/texas-teen-dwi-wreck/index.html'
...

>>> print(cnn_paper.size()) # cnn has 3100 articles
3100

Article caching¶

By default, newspaper caches all previously extracted articles and eliminates any article which it has already extracted.

This feature exists to prevent duplicate articles and to increase extraction speed.

>>> cbs_paper = newspaper.build('http://cbs.com')
>>> cbs_paper.size()
1030

>>> cbs_paper = newspaper.build('http://cbs.com')
>>> cbs_paper.size()
2

The return value of cbs_paper.size() changes from 1030 to 2 because when we first crawled cbs we found 1030 articles. However, on our second crawl, we eliminate all articles which have already been crawled.

This means 2 new articles have been published since our first extraction.

You may opt out of this feature with the memoize_articles parameter.

You may also pass in the lower level``Config`` objects as covered in the advanced section.

>>> import newspaper

>>> cbs_paper = newspaper.build('http://cbs.com', memoize_articles=False)
>>> cbs_paper.size()
1030

>>> cbs_paper = newspaper.build('http://cbs.com', memoize_articles=False)
>>> cbs_paper.size()
1030

Extracting Source categories¶

>>> for category in cnn_paper.category_urls():
>>>     print(category)

u'http://lifestyle.cnn.com'
u'http://cnn.com/world'
u'http://tech.cnn.com'
...

Extracting Source feeds¶

>>> for feed_url in cnn_paper.feed_urls():
>>>     print(feed_url)

u'http://rss.cnn.com/rss/cnn_crime.rss'
u'http://rss.cnn.com/rss/cnn_tech.rss'
...

Extracting Source brand & description¶

>>> print(cnn_paper.brand)
u'cnn'

>>> print(cnn_paper.description)
u'CNN.com delivers the latest breaking news and information on the latest...'

News Articles¶

Article objects are abstractions of news articles. For example, a news Source would be CNN while a news Article would be a specific CNN article. You may reference an Article from an existing news Source or initialize one by itself.

Referencing it from a Source.

>>> first_article = cnn_paper.articles[0]

Initializing an Article by itself.

>>> from newspaper import Article
>>> first_article = Article(url="http://www.lemonde.fr/...", language='fr')

Note the similar language= named paramater above. All the config parameters as described for Source objects also apply for Article objects! Source and Article objects have a very similar api.

There are endless possibilities on how we can manipulate and build articles.

Downloading an Article¶

We begin by calling download() on an article. If you are interested in how to quickly download articles concurrently with multi-threading check out the advanced section.

>>> first_article = cnn_paper.articles[0]

>>> first_article.download()

>>> print(first_article.html)
u'<!DOCTYPE HTML><html itemscope itemtype="http://...'

>>> print(cnn_paper.articles[7].html)
u'' fail, not downloaded yet

Parsing an Article¶

You may also extract meaningful content from the html, like authors and body-text. You must have called download() on an article before calling parse().

>>> first_article.parse()

>>> print(first_article.text)
u'Three sisters who were imprisoned for possibly...'

>>> print(first_article.top_image)
u'http://some.cdn.com/3424hfd4565sdfgdg436/

>>> print(first_article.authors)
[u'Eliott C. McLaughlin', u'Some CoAuthor']

>>> print(first_article.title)
u'Police: 3 sisters imprisoned in Tucson home'

>>> print(first_article.images)
['url_to_img_1', 'url_to_img_2', 'url_to_img_3', ...]

>>> print(first_article.movies)
['url_to_youtube_link_1', ...] # youtube, vimeo, etc

Performing NLP on an Article¶

Finally, you may extract out natural language properties from the text. You must have called both download() and parse() on the article before calling nlp().

As of the current build, nlp() features only work on western languages.

>>> first_article.nlp()

>>> print(first_article.summary)
u'...imprisoned for possibly a constant barrage...'

>>> print(first_article.keywords)
[u'music', u'Tucson', ... ]

>>> print(cnn_paper.articles[100].nlp()) # fail, not been downloaded yet
Traceback (...
ArticleException: You must parse an article before you try to..

nlp() is expensive, as is parse(), make sure you actually need them before calling them on all of your articles! In some cases, if you just need urls, even download() is not necessary.

Easter Eggs¶

Here are random but hopefully useful features! hot() returns a list of the top trending terms on Google using a public api. popular_urls() returns a list of popular news source urls.. In case you need help choosing a news source!

>>> import newspaper

>>> newspaper.hot()
['Ned Vizzini', Brian Boitano', Crossword Inventor', 'Alex & Sierra', ... ]

>>> newspaper.popular_urls()
['http://slate.com', 'http://cnn.com', 'http://huffingtonpost.com', ... ]

>>> newspaper.languages()

Your available languages are:
input code      full name

  ar              Arabic
  de              German
  en              English
  es              Spanish
  fr              French
  he              Hebrew
  it              Italian
  ko              Korean
  no              Norwegian
  pt              Portuguese
  sv              Swedish
  zh              Chinese

Advanced¶

This section of the docs shows how to do some useful but advanced things with newspaper.

Multi-threading article downloads¶

Downloading articles one at a time is slow. But spamming a single news source like cnn.com with tons of threads or with ASYNC-IO will cause rate limiting and also doing that is very mean.

We solve this problem by allocating 1-2 threads per news source to both greatly speed up the download time while being respectful.

>>> import newspaper
>>> from newspaper import news_pool

>>> slate_paper = newspaper.build('http://slate.com')
>>> tc_paper = newspaper.build('http://techcrunch.com')
>>> espn_paper = newspaper.build('http://espn.com')

>>> papers = [slate_paper, tc_paper, espn_paper]
>>> news_pool.set(papers, threads_per_source=2) # (3*2) = 6 threads total
>>> news_pool.join()

At this point, you can safely assume that download() has been
called on every single article for all 3 sources.

>>> print(slate_paper.articles[10].html)
u'<html> ...'

Keeping Html of main body article¶

Keeping the html of just an article’s body text is helpbut because it allows you to retain some of the semantic information in the html. Also it will help if you end up displaying the extracted article somehow.

Here is how to do so:

>>> from newspaper import Article

>>> a = Article('http://www.cnn.com/2014/01/12/world/asia/north-korea-charles-smith/index.html'
    , keep_article_html=True)

>>> a.download()
>>> a.parse()

>>> a.article_html
u'<div> \n<p><strong>(CNN)</strong> -- Charles Smith insisted Sunda...'

The lxml (dom object) and top_node (chunk of dom that contains our ‘Article’) are also cached incase users would like to use them.

Access after parsing() with:

>>> a.download()
>>> a.parse()
>>> a.clean_dom
<lxml object ...  >

>>> a.clean_top_node
<lxml object ...  >

Adding new languages¶

First, please reference this file and read from the highlighted line all the way down to the end of the file.

https://github.com/codelucas/newspaper/blob/master/newspaper/text.py#L57

One aspect of our text extraction algorithm revolves around counting the number of stopwords present in a text. Stopwords are: some of the most common, short function words, such as the, is, at, which, and on in a language.

Reference this line to see it in action: https://github.com/codelucas/newspaper/blob/master/newspaper/extractors.py#L668

So for latin languages, it is pretty basic. We first provide a list of stopwords in stopwords-<language-code>.txt form. We then take some input text and tokenize it into words by splitting the white space. After that we perform some bookkeeping and then proceed to count the number of stopwords present.

For non-latin languages, as you may have noticed in the code above, we need to tokenize the words in a different way, splitting by whitespace simply won’t work for languages like Chinese or Arabic. For the Chinese language we are using a whole new open source library called jieba to split the text into words. For arabic we are using a special nltk tokenizer to do the same job.

So, to add full text extraction to a new (non-latin) language, we need:

1. Push up a stopwords file in the format of stopwords-<2-char-language-code>.txt in newspaper/resources/text/.

2. Provide a way of splitting/tokenizing text in that foreign language into words. Here are some examples for Chinese, Arabic, English

For latin languages:

1. Push up a stopwords file in the format of stopwords-<2-char-language-code>.txt in newspaper/resources/text/. and we are done!

Explicitly building a news source¶

Instead of using the newspaper.build(..) api, we can take one step lower into newspaper’s Source api.

>>> from newspaper import Source
>>> cnn_paper = Source('http://cnn.com')

>>> print(cnn_paper.size()) # no articles, we have not built the source
0

>>> cnn_paper.build()
>>> print(cnn_paper.size())
3100

Note the build() method above. You may go lower level and de-abstract it for absolute control over how your sources are constructed.

>>> cnn_paper = Source('http://cnn.com')
>>> cnn_paper.download()
>>> cnn_paper.parse()
>>> cnn_paper.set_categories()
>>> cnn_paper.download_categories()
>>> cnn_paper.parse_categories()
>>> cnn_paper.set_feeds()
>>> cnn_paper.download_feeds()
>>> cnn_paper.generate_articles()

>>> print(cnn_paper.size())
3100

And voila, we have mimic’d the build() method. In the above sequence, every method is dependant on the method above it. Stop whenever you wish.

Parameters and Configurations¶

Newspaper provides two api’s for users to configure their Article and Source objects. One is via named parameter passing recommended and the other is via Config objects.

Here are some named parameter passing examples:

>>> import newspaper
>>> from newspaper import Article, Source

>>> cnn = newspaper.build('http://cnn.com', language='en', memoize_articles=False)

>>> article = Article(url='http://cnn.com/french/...', language='fr', fetch_images=False)

>>> cnn = Source(url='http://latino.cnn.com/...', language='es', request_timeout=10,
                                                            number_threads=20)

Here are some examples of how Config objects are passed.

>>> import newspaper
>>> from newspaper import Config, Article, Source

>>> config = Config()
>>> config.memoize_articles = False

>>> cbs_paper = newspaper.build('http://cbs.com', config)

>>> article_1 = Article(url='http://espn/2013/09/...', config)

>>> cbs_paper = Source('http://cbs.com', config)

Here is a full list of the configuration options:

keep_article_html, default False, “set to True if you want to preserve html of body text”

http_success_only, default True, “set to False to capture non 2XX responses as well”

MIN_WORD_COUNT, default 300, “num of word tokens in article text”

MIN_SENT_COUNT, default 7, “num of sentence tokens”

MAX_TITLE, default 200, “num of chars in article title”

MAX_TEXT, default 100000, “num of chars in article text”

MAX_KEYWORDS, default 35, “num of keywords in article”

MAX_AUTHORS, default 10, “num of author names in article”

MAX_SUMMARY, default 5000, “num of chars of the summary”

MAX_FILE_MEMO, default 20000, “python setup.py sdist bdist_wininst upload”

memoize_articles, default True, “cache and save articles run after run”

fetch_images, default True, “set this to false if you don’t care about getting images”

image_dimension_ration, default 16/9.0, “max ratio for height/width, we ignore if greater”

language, default ‘en’, “run newspaper.languages() to see available options.”

browser_user_agent, default ‘newspaper/%s’ % __version__

request_timeout, default 7

number_threads, default 10, “number of threads when mthreading”

verbose, default False, “turn this on when debugging”

You may notice other config options in the newspaper/configuration.py file, however, they are private, please do not toggle them.

Caching¶

TODO

Specifications¶

Here, we will define exactly how newspaper handles a lot of the data extraction.

TODO

Contributors¶

Maintained and authored by:¶

Lucas Ou-Yang – http://codelucas.com, lucasyangpersonal@gmail.com

Thanks to the following contributors:¶

https://github.com/codelucas/newspaper/graphs/contributors

Newspaper relied on some code of a few other open source projects:¶

Thanks to all who have contributed to python-goose. You can find the contributors list here: https://github.com/grangier/python-goose/graphs/contributors

Thanks to all who have contributed to PyTeaser. You can find the contributors list here: https://github.com/xiaoxu193/PyTeaser/graphs/contributors

Thanks to all who have contributed to gravity-goose. You can find the contributors list here: https://github.com/GravityLabs/goose/graphs/contributors

Thanks to all who have contributed to jieba. You can find the contributors list here: https://github.com/fxsjy/jieba/graphs/contributors

Thanks to all who have contributed to nltk. You can find the contributors list here: https://github.com/nltk/nltk/graphs/contributors

Thanks to all who have contributed to lxml. You can find the contributors list here: http://lxml.de/credits.html

Thanks to all who have contributed to requests. You can find the contributors list here: https://github.com/kennethreitz/requests/graphs/contributors

LICENSE¶

Authored and maintained by Lucas Ou-Yang.

Newspaper uses a lot of python-goose’s parsing code. View their license here.

Please feel free to email & contact me if you run into issues or just would like to talk about the future of this library and news extraction in general!