html-text 0.4.1

Extract text from HTML

HTML to Text

Extract text from HTML

• Free software: MIT license

How is html_text different from .xpath('//text()') from LXML or .get_text() from Beautiful Soup? Text extracted with html_text does not contain inline styles, javascript, comments and other text that is not normally visible to the users. It normalizes whitespace, but is also smarter than .xpath('normalize-space()), adding spaces around inline elements (which are often used as block elements in html markup), tries to avoid adding extra spaces for punctuation and can add newlines so that the output text looks like how it is rendered in browsers.

Apart from just getting text from the page (e.g. for display or search), one intended usage of this library is for machine learning (feature extraction). If you want to use the text of the html page as a feature (e.g. for classification), this library gives you plain text that you can later feed into a standard text classification pipeline. If you feel that you need html structure as well, check out webstruct library.

Install

Install with pip:

pip install html-text

The package depends on lxml, so you might need to install some additional packages: http://lxml.de/installation.html

Usage

Extract text from HTML:

>>> import html_text
>>> html_text.extract_text('<h1>Hello</h1> world!')
'Hello\n\nworld!'

>>> html_text.extract_text('<h1>Hello</h1> world!', guess_layout=False)
'Hello world!'

You can also pass already parsed lxml.html.HtmlElement:

>>> import html_text
>>> tree = html_text.parse_html('<h1>Hello</h1> world!')
>>> html_text.extract_text(tree)
'Hello\n\nworld!'

Or define a selector to extract text only from specific elements:

>>> import html_text
>>> sel = html_text.cleaned_selector('<h1>Hello</h1> world!')
>>> subsel = sel.xpath('//h1')
>>> html_text.selector_to_text(subsel)
'Hello'

Passed html will be first cleaned from invisible non-text content such as styles, and then text would be extracted. NB Selectors are not cleaned automatically you need to call html_text.cleaned_selector first.

Main functions:

• html_text.extract_text accepts html and returns extracted text.

• html_text.cleaned_selector accepts html as text or as lxml.html.HtmlElement, and returns cleaned parsel.Selector.

• html_text.selector_to_text accepts parsel.Selector and returns extracted text.

If guess_layout is True (default), a newline is added before and after newline_tags, and two newlines are added before and after double_newline_tags. This heuristic makes the extracted text more similar to how it is rendered in the browser. Default newline and double newline tags can be found in html_text.NEWLINE_TAGS and html_text.DOUBLE_NEWLINE_TAGS.

It is possible to customize how newlines are added, using newline_tags and double_newline_tags arguments (which are html_text.NEWLINE_TAGS and html_text.DOUBLE_NEWLINE_TAGS by default). For example, don’t add a newline after <div> tags:

>>> newline_tags = html_text.NEWLINE_TAGS - {'div'}
>>> html_text.extract_text('<div>Hello</div> world!',
...                        newline_tags=newline_tags)
'Hello world!'

Credits

The code is extracted from utilities used in several projects, written by Mikhail Korobov.

---

History

0.4.1 (2018-09-25)

Fixed a regression in 0.4.0 release: text was empty when html_text.extract_text is called with a node with text, but without children.

0.4.0 (2018-09-25)

This is a backwards-incompatible release: by default html_text functions now add newlines after elements, if appropriate, to make the extracted text to look more like how it is rendered in a browser.

To turn it off, pass guess_layout=False option to html_text functions.

• guess_layout option to to make extracted text look more like how it is rendered in browser.

• Add tests of layout extraction for real webpages.

0.3.0 (2017-10-12)

• Expose functions that operate on selectors, use .//text() to extract text from selector.

0.2.1 (2017-05-29)

• Packaging fix (include CHANGES.rst)

0.2.0 (2017-05-29)

• Fix unwanted joins of words with inline tags: spaces are added for inline tags too, but a heuristic is used to preserve punctuation without extra spaces.

• Accept parsed html trees.

0.1.1 (2017-01-16)

• Travis-CI and codecov.io integrations added

0.1.0 (2016-09-27)

• First release on PyPI.