a fast and platform independent readability port (JS)
a fast and platform independent readability port
This is a port of the algorithm used by the Readability bookmarklet to extract relevant pieces of information from websites, using a SAX parser.
The advantage over other ports, e.g. arrix/node-readability, is a smaller memory footprint and a much faster execution. In my tests, most pages, even large ones, were finished within 15ms (on node, see below for more information). It works with Rhino, so it runs on YQL, which may have interesting uses. And it works within a browser.
The Readability extraction algorithm was completely ported, but some adjustments were made:
tags are recognized and gain a higher value
If a heading is part of the pages, it is removed (Readability removed any single
, and ignored other tags)
instapaper-bodyare classes to show an algorithm like this where the content is. readabilitySAX recognizes them and adds additional points
Every bit of code that was taken from the original algorithm was optimized, eg. RegExps should now perform faster (they were optimized & use
String#match, which doesn't force the interpreter to build an array)
Some improvements made by GGReadability (an Obj-C port of Readability) were adopted
widthattributes are high - icon sized images (<= 32px) get skipped
npm install readabilitySAX
A command line interface (CLI) may be installed via
npm install -g readabilitySAX
It's then available via
To get this readme, just run
The format is optional (it's either
html, the default value is
require("readabilitySAX"). You'll get an object containing three methods:
Readability(settings): The readability constructor. It works as a handler for
htmlparser2. Read more about it in the wiki!
WritableStream(settings, cb): A constructor that unites
Readabilityconstructor. It's a writable stream, so simply
.writeall your data to it. Your callback will be called once
.endwas called. Bonus: You can also
.pipedata into it!
createWritableStream(settings, cb): Returns a new instance of the
WritableStream. (It's a simple factory method.)
There are two methods available that are deprecated and will be removed in a future version:
get(link, [settings], callback): Gets a webpage and process it.
process(data): Takes a string, runs readabilitySAX and returns the page.
Please don't use those two methods anymore. Streams are the way you should build interfaces in node, and that's what I want encourage people to use.
I started to implement simplified SAX-"parsers" for Rhino/YQL (using E4X) and the browser (using the DOM) to increase the overall performance on those platforms. The DOM version is inside the
A demo of how to use readabilitySAX inside a browser may be found at jsFiddle. Some basic example files are inside the
A table using E4X-based events is available as the community table
redabilitySAX, as well as here.
Most SAX parsers (as sax.js) fail when a document is malformed XML, even if it's correct HTML. readabilitySAX should be used with htmlparser2, my fork of the
htmlparser-module (used by eg.
jsdom), which corrects most faults. It's listed as a dependency, so npm should install it with readabilitySAX.
Using a package of 724 pages from CleanEval (their website seems to be down, try to google it), readabilitySAX processed all of them in 5768 ms, that's an average of 7.97 ms per page.
The benchmark was done using
tests/benchmark.json a MacBook (late 2010) and is probably far from perfect.
Performance is the main goal of this project. The current speed should be good enough to run readabilitySAX on a singe-threaded web server with an average number of requests. That's an accomplishment!
The main goal of CleanEval is to evaluate the accuracy of an algorithm.