Port of Readability HTML parser in Python


Keywords
bookie, breadability, content, HTML, parsing, readability, readable, python, text-mining, text-extraction, html-parsing, html-extraction, html-extractor
License
BSD-3-Clause
Install
pip install breadability==0.1.14

Documentation

breadability - another readability Python (v2.6-v3.3) port

https://api.travis-ci.org/bookieio/breadability.png?branch=master

I've tried to work with the various forks of some ancient codebase that ported readability to Python. The lack of tests, unused regex's, and commented out sections of code in other Python ports just drove me nuts.

I put forth an effort to bring in several of the better forks into one code base, but they've diverged so much that I just can't work with it.

So what's any sane person to do? Re-port it with my own repo, add some tests, infrastructure, and try to make this port better. OSS FTW (and yea, NIH FML, but oh well I did try)

This is a pretty straight port of the JS here:

Alternatives

Installation

This does depend on lxml so you'll need some C headers in order to install things from pip so that it can compile.

$ [sudo] apt-get install libxml2-dev libxslt-dev
$ [sudo] pip install git+git://github.com/bookieio/breadability.git

Tests

$ nosetests-2.6 tests && nosetests-3.2 tests && nosetests-2.7 tests && nosetests-3.3 tests

Usage

Command line

$ breadability http://wiki.python.org/moin/BeginnersGuide

Options

  • b will write out the parsed content to a temp file and open it in a browser for viewing.
  • d will write out debug scoring statements to help track why a node was chosen as the document and why some nodes were removed from the final product.
  • f will override the default behaviour of getting an html fragment (<div>) and give you back a full <html> document.
  • v will output in verbose debug mode and help let you know why it parsed how it did.

Python API

from __future__ import print_function

from breadability.readable import Article


if __name__ == "__main__":
    document = Article(html_as_text, url=source_url)
    print(document.readable)

Work to be done

Yep, I've got some catching up to do. I don't do pagination, I've got a lot of custom tweaks I need to get going, there are some articles that fail to parse. I also have more tests to write on a lot of the cleaning helpers, but hopefully things are setup in a way that those can/will be added.

Fortunately, I need this library for my tools:

so I really need this to be an active and improving project.

Off the top of my heads TODO list:

  • Support metadata from parsed article [url, confidence scores, all candidates we thought about?]
  • More tests, more thorough tests
  • More sample articles we need to test against in the test_articles
  • Tests that run through and check for regressions of the test_articles
  • Tidy'ing the HTML that comes out, might help with regression tests ^^
  • Multiple page articles
  • Performance tuning, we do a lot of looping and re-drop some nodes that should be skipped. We should have a set of regression tests for this so that if we implement a change that blows up performance we know it right away.
  • More docs for things, but sphinx docs and in code comments to help understand wtf we're doing and why. That's the biggest hurdle to some of this stuff.

Inspiration