Update: A lot of things happened since the publication of this article. First of all, I have updated this article with HPricot and scRUBYt! examples – then I wrote the second part, I hacked up a Ruby web-scraping toolkit, scRUBYt! which also has a community web page – check it out, it’s hot right now!
Introduction
Despite of the ongoing Web 2.0 buzz, the absolute majority of the Web pages
are still very Web 1.0: They heavily mix presentation with content.
[1] This makes hard or impossible for a computer to tell
off the wheat from the chaff: to sift out meaningful data from the rest of the elements
used for formatting, spacing, decoration or site navigation.
To remedy this problem, some sites provide access to their content
through APIs (typically via web services), but in practice nowadays this is
limited to a few (big) sites, and some of them are not even free or public.
In an ideal Web 2.0 world, where data sharing and site interoperability is one of
the basic principles, this should change soon(?) – but what should
one do if he needs the data NOW and not in the likely-to-happen-future?
Manic Miner
The solution is called screen/Web scraping or Web extraction – mining Web data
by observing the page structure and wrapping out the relevant records. In some
cases the task is even more complex than that: The data can be scattered over
more pages, triggering of a GET/POST request may be needed to get the input page
for the extraction or authorization may be required to navigate to the page of
interest. Ruby has solutions for these issues, too – we will take a look at them
as well.
The extracted data can be used in any way you like – to create mashups
(e.g. chicagocrime.org by Django author
Adrian Holovaty), to remix and present the relevant data
(e.g. rubystuff.com com by
ruby-doc.org maintainer James Britt), to automatize
processes (for example if you have more bank accounts, to get the sum of the
money you have all together, without using your browser), monitor/compare
prices/items, meta-search, create a semantic web page out of a regular one –
just to name a few. The number of the possibilities is limited by your
imagination only.
Tools of the trade
In this section we will check out the two main possibilities (string and tree based
wrappers) and take a look at HTree, REXML, RubyfulSoup and WWW::Mechanize based
solutions.
String wrappers
The easiest (but in most of the cases inadequate) possibility is to view the
HTML document as a string. In this case you can use regular expressions to
mine the relevant data. For example if you would like to extract names
of goods and their price from a Web shop, and you know that they are
both in the same HTML element, like:
<td>Samsung SyncMasta 21''LCD $750.00</td>
you can extract this record from Ruby with this code snippet:
scan(page, /<td>(.*)\s+(\$\d+\.\d{2})<\/td>/)
Let’s see a real (although simple) example:
1 require 'open-uri' 2 url = "http://www.google.com/search?q=ruby" 3 open(url) { 4 |page| page_content = page.read() 5 links = page_content.scan(/<a class=l.*?href=\"(.*?)\"/).flatten 6 links.each {|link| puts link} 7 }
The first and crucial part of creating the wrapper program was the observation of the
page source: We had to look for something that appears only in the result links.
In this case this was the presence of the ‘class’ attribute, with value ‘l’. This
task is usually not this easy, but for illustration purposes it serves well.
This minimalistic example shows the basic concepts: How to load the
contents of a Web page into a string (line 4), and how to extract the result
links on a google search result page (line 5). (After execution, the program
will list the first 10 links of a google search query for the word ‘ruby’ (line 6)).
However, in practice you will mostly need to extract data which are not
in a contiguous string, but contained in multiple HTML tags, or divided
in a way where a string is not the proper structure for searching. In
this case it is better to view the HTML document as a tree.[2]
Tree wrappers
The tree-based approach, although enables more powerful techniques,
has its problems, too: The HTML document can look very good in a browser,
yet still be seriously malformed (unclosed/misused tags). It is a
non-trivial problem to parse such a document into a structured format
like XML, since XML parsers can work with well-formed documents only.
HTree and REXML
There is a solution (in most of the cases) for this problem, too:
It is called HTree. This handy package is able
to tidy up the malformed HTML input, turning it to XML – the recent version is
capable to transform the input into the nicest possible XML from our point of view: a REXML
Document. (
REXML is Ruby’s standard XML/XPath processing library).
After preprocessing the page content with HTree, you can unleash the
full power of XPath, which is a very powerful XML document querying language,
highly suitable for web extraction.
Refer to [3] for the installation instructions of HTree.
Let’s revisit the previous Google example:
1 require 'open-uri' 2 require 'htree' 3 require 'rexml/document' 4 url = "http://www.google.com/search?q=ruby" 5 open(url) { 6 |page| page_content = page.read() 7 doc = HTree(page_content).to_rexml 8 doc.root.each_element('//a[@class="l"]') { |elem| puts elem.attribute('href').value } 9 }
HTree is used in the 7th line only – it converts the HTML page (loaded into the pageContent
variable on the previous line) into a REXML Document. The real magic happens
in the 8th line. We select all the <a> tags which have an attribute ‘class’ with the
value ‘l’, then for each such element write out the ‘href’ attribute. [4]
I think this approach is much more natural for querying an XML document than a regular expression.
The only drawback is that you have to learn a new language, XPath, which is (mainly from
version 2.0) quite difficult to master. However, just to get started you do not need to know
much of it, yet you gain lots of raw power compared to the possibilities offered by regular expressions.
Hpricot
Hpricot is “a Fast, Enjoyable HTML Parser for Ruby” by one of the coolest (Ruby) programmers of our century, why the lucky stiff. From my experience, the tag line is absolutely correct – Hpricot is both very fast (thanks to a C based scanner implementation) and really fun to use.
It is based on HTree and JQuery, thus it can provide the same functionality as the previous Htree + REXML combination, but with a much better performance and greater ease of use. Let’s see the google example again – I guess you will understand instantly what I mean!
1 require 'rubygems' 2 require 'hpricot' 3 require 'open-uri' 4 doc = Hpricot(open('http://www.google.com/search?q=ruby')) 5 links = doc/"//a[@class=l]" 6 links.map.each {|link| puts link.attributes['href']}
Well, though this was slightly easier than with the tools seen so far, this example does not really show the power of Hpricot – there is much, much, much more in the store: different kinds of parsing, CSS selectors and searches, nearly full XPath support, and lots of chunky bacon! If you are doing something smaller and don’t need the power of scRUBYt!, my advice is to definitely use Hpricot from the tools listed here. For more information, installation instructions, tutorials and documentation check out Hpricot’ s homepage!
RubyfulSoup
Rubyfulsoup is a very powerful Ruby
screen-scraping package, which offers
similar possibilities like HTree + XPath. For people who are not handy with XML/XPath,
RubyfulSoup may be a wise compromise: It’s an all-in-one, effective HTML parsing
and web scraping tool with Ruby-like syntax. Although it’s expressive power
lags behind XPath2.0, it should be adequate in 90% of the cases. If your problem is in the
remaining 10%, you probably don’t need to read this tutorial anyway 😉
Installation instructions can be found here: [5].
The google example again:
1 require 'rubygems' 2 require 'rubyful_soup' 3 require 'open-uri' 4 url = "http://www.google.com/search?q=ruby" 5 open(url) { 6 |page| page_content = page.read() 7 soup = BeautifulSoup.new(page_content) 8 result = soup.find_all('a', :attrs => {'class' => 'l'}) 9 result.each { |tag| puts tag['href'] } 10 }
As you can see, the difference between the HTree + REXML and RubyfulSoup examples is minimal –
basically it is limited to differences in the querying syntax. On line 8, you look up all the
<a> tags, with the specified attribute list (in this case a hash with a single pair { ‘class’ => ‘l’ } )
The other syntactical difference is looking up the value of the ‘href’ attribute on line 9.
I have found RubyfulSoup the ideal tool for screen scraping from a single page – however web navigation
(GET/POST, authentication, following links) is not really possible or obscure at best with
this tool (which is perfectly OK, since it does not aim to provide this functionality). However, there
is nothing to fear – the next package is doing just exactly that.
WWW::Mechanize
As of today, prevalent majority of data resides in the deep Web – databases, that
are accessible via querying through web-forms. For example if you would like to get information
on flights from New York to Chicago, you will (hopefully) not search for it on google –
you go to the website of the Ruby Airlines instead, fill in the adequate fields and click on search.
The information which appears is not available on a static page – it’s looked up on demand and
generated on the fly – so until the very moment the web server generates it for you , its practically
non-existent (i.e. it resides in the deep Web) and hence impossible to extract. At this point
WWW::Mecahnize comes into play.
(See [6] for installation instructions)
WWW::Mechanize belongs to the family of screen scraping products (along with http-access2 and Watir)
that are capable to drive a browser. Let’s apply the ‘Show, don’t tell’ mantra – for everybody’s delight
and surprise, illustrated on our google scenario:
require 'rubygems' require 'mechanize' agent = WWW::Mechanize.new page = agent.get('http://www.google.com') search_form = page.forms.with.name("f").first search_form.fields.name("q").first.value = "ruby" search_results = agent.submit(search_form) search_results.links.each { |link| puts link.href if link.class_name == "l" }
I have to admit that i have been cheating with this one ;-). I had to hack WWW::Mechanize to
access a custom attribute (in this case ‘class’) because normally this is not available.
See how i did it here: [7]
This example illustrates a major difference between RubyfulSoup and Mechanize: additionally to screen scraping
functionality, WWW::mechanize is able to drive the web browser like a human user: It filled in the
search form and clicked the ‘search’ button, navigating to the result page, then performed screen scraping
on the results.
This example also pointed out the fact that RubyfulSoup – although lacking navigation possibilities –
is much more powerful in screen scraping. For example, as of now, you can not extract arbitrary (say <p>)
tags with Mechanize, and as the example illustrated, attribute extraction is not possible either – not to
mention more complex, XPath like queries (e.g. the third <td> in the second <tr>) which is easy with
RubyfulSoup/REXML. My recommendation is to combine these tools, as pointed out in the last section of this article.
scRUBYt!
scRUBYt! is a simple to learn and use, yet very powerful web extraction framework written in Ruby, based on Hpricot and Mechanize. Well, yeah, I made it 🙂 so this is kind of a self promotion, but I think (hopefully not just because being overly biased ;-)) it is the most powerful web extraction toolkit available to date. scRUBYt! can navigate through the Web (like clicking links, filling textfields, crawling to further pages – thanks to mechanize), extract, query, transform and save relevant data from the Web page of your interest by the concise and easy to use DSL (thanks to Hpricot and a lots of smart heuristics).
OK, enough talking – let’s see it in action! I guess this is rather annoying now for the 6th time, but let’s revisit the google example once more! (for the last time, I promise 🙂
1 require 'rubygems' 2 require 'scrubyt' 3 google_data = Scrubyt::Extractor.define do 4 fetch 'http://www.google.com/ncr' 5 fill_textfield 'q', 'ruby' 6 submit 7 result 'Ruby Programming Language' do 8 link 'href', :type => :attribute 9 end 10 end 11 google_data.to_xml.write($stdout, 1) 12 Scrubyt::ResultDumper.print_statistics(google_data)
Oputput:
<root> <result> <link>http://www.ruby-lang.org/</link> </result> <result> <link>http://www.ruby-lang.org/en/20020101.html</link> </result> <result> <link>http://en.wikipedia.org/wiki/Ruby_programming_language</link> </result> <result> <link>http://en.wikipedia.org/wiki/Ruby</link> </result> <result> <link>http://www.rubyonrails.org/</link> </result> <result> <link>http://www.rubycentral.com/</link> </result> <result> <link>http://www.rubycentral.com/book/</link> </result> <result> <link>http://www.w3.org/TR/ruby/</link> </result> <result> <link>http://poignantguide.net/</link> </result> <result> <link>http://www.zenspider.com/Languages/Ruby/QuickRef.html</link> </result> </root> result extracted 10 instances. link extracted 10 instances.
You can donwload this example from here.
Though the code snippet is not really shorter, maybe even longer than the other ones, there are a lots of thing to note here: First of all, instead of loading the page directly (you can do that as well, of course), scRUBYt allows you to navigate there by going to google, filling the appropriate text field and submitting the search. The next interesting thing is that you need no XPaths or other mechanism to query your data – you just copy’n’ paste some examples from the page, and that’s it. Also, the whole description of the scraping process is more human friendly – you do not need to care about URLs, HTML, passing the document around, handling the result – everything is hidden from you and controlled by scRUBYt!’s DSL instead. You even get a nice statistics on how much stuff was extracted. 🙂
The above example is just the top of the iceberg – there is much, much, much more in scRUBYt! than what you have seen so far. If you would like to know more, check out the tutorials and other goodies on scRUBYt!’s homepage.
WATIR
From the WATIR page:
WATIR stands for “Web Application Testing in Ruby”. Watir drives the Internet Explorer browser the same
way people do. It clicks links, fills in forms, presses buttons. Watir also checks results, such as whether
expected text appears on the page.
Unfortunately I have no experience with WATIR since i am a linux-only nerd, using windows for occasional
gaming but not for development, so I can not tell anything about it from the first hand, but judging from the
mailing list contributions i think Watir is more mature and feature-rich than mechanize. Definitely
check it out if you are running on Win32.
The silver bullet
For a complex scenario, usually an amalgam of the above tools can provide the ultimate solution:
The combination of WWW::Mechanize or WATIR (for automatization of site navigation), RubyfulSoup (for serious screen
scraping, where the above two are not enough) and HTree+REXML (for extreme cases where even RubyfulSoup
can’t help you).
I have been creating industry-strength, robust and effective screen scraping solutions in the last five years
of my career, and i can show you a handful of pages where even the most sophisticated solutions do not work (and
i am not talking about scraping with RubyfulSoup here, but even more powerful solutions (like embedding
mozilla in your application and directly accessing the DOM etc)). So the basic rule is: there is no
spoon (err… silver bullet) – and i know by experience that the number of ‘hard-to-scrap’ sites is rising
(partially because of the Web 2.0 stuff like AJAX, but also because some people would not like their sites to
be extracted and apply different anti-scraping masquerading techniques).
The described tools should be enough to get you started – additionally, you may have to figure out how to
drill down to your stuff on the concrete page of interest.
In the next installment of this series, i will create a mashup application using the introduced tools, from some
more interesting data than google 😉
The results will be presented on a Ruby on Rails powered page, in a sortable AJAX table.
If you liked the article, subscribe to the rubyrailways.com feed!

etc.), but these are falling out of scope of the current topic.Back
to use them for several reasons: No additional packages are needed (this is even more important if you don’t have
install rights), you don’t have to rely on the HTML parser’s output and if you can use regular expressions, it’s
usually the easier way to do so. Back
wget http://cvs.m17n.org/viewcvs/ruby/htree.tar.gz (or download it from your browser)
tar -xzvf htree.tar.gz
sudo ruby install.rb Back
each_element_with_attribute, or a different, more effective XPath – I have chosen to use
this method to get as close to the regexp example as possible, so it is easy to observe
the difference between the two approaches for the same solution. For a real REXML tutorial/documentation
visit the REXML site.
Back
sudo gem install rubyful_soup
Since it was installed as a gem, don’t forget to require ‘rubygems’ before requiring rubyful_soup.
Back
Back
To the class definition:
attr_reader :class_name
Into the constructor:
@class_name = node.attributes['class']
Pingback: Ruby, Rails, Web2.0 » Blog Archive » Announcing screen-scraping series
There is a project to port Watir for Firefox, just FYI – it’s called FireWatir
http://wiki.mozilla.org/SoftwareTesting:WatirandFirefox
those were supposed to be underscores around the and, between Watir and Firefox, in the url in my above comment – I don’t know how those got altered – sorry.
Chris,
Thanks for the link! We are developing a screen scraping application just now which is a Firefox extension, so i am quite involved with Firefox and good to know about stuff like FireWatir.
About the underscores – i guess it is WordPress. For example this was written as asterisk-this-asterisk and now you can see it in bold. Probably undescore is a shortcut for italic i guess…
I am not commenting on your blog because you had a captcha-esque ‘please add 10 and 0’ field, it derided me as ‘not knowning math’ when I entered “10”, and it eradicated all the contents of my post rather than letting me take the challenge again. Comment spam is annoying, but my time is better spent bitching about your way of handling it than actually rewriting my post and helping you out.
@aa:
😉 Sorry for the inconvenience…. Nobody ‘bitched’ about it yet, so I did not know (I have tried it once or twice and it worked OK for me). What do you suggest? I would not like to drop the captcha completely since then i am receiving LOTS of spam. Maybe somebody has a suggestion for a better system?
OK, I have turned off the captcha until i find something more convenient… So if the comments will be full of spam its because of that 😉
With WWW::Mechanize you can get the parsed rexml document and it also adds convenience methods to this REXML::Document
agent = WWW::Mechanize.new
page = agent.get(‘http://www.google.com’)
form = page.forms.first
form.fields.name(‘q’).value = ‘ruby’
searchresults = agent.submit(form)
searchresults.root.each_element(‘//a[@class=”l”]’) {|elem| puts elem.attribute(‘href’).value }
Absolutely superb article. I’ve generally always put up with using old fashioned regexp in my screen scraping and didn’t know of these other methods until now. You’ve opened my eyes. Thank you.
Pingback: Pig Pen - Web Standards Compliant Web Design Blog » Blog Archive » Screen Scraping With Ruby
Super neat. I’m expecting a new one, specially using Gecko’s DOM.
Great introduction. You might want to add an link to your main page in the posts. I tried clicking the header but there seems to be no link. Now I will have to go back to Reddit to find out your blog’s url (I’m using sharpreader).
@Leonardo:
I have a fully working and tested Java solution for that – but there i have every building stone ( Java gecko widget – currently using SWT.Browser but there are alternatives like Ajax Toolkit Framework and XULRunner which are even better) and JavaXPCOM + W3CConnector to communicate between mozilla and java)
The problem with Ruby is that although both of these things are there (RubyGecko, GTK::Mozembed) and rbXPCOM, they are in a very-very immature state, i am not sure if even usable. So although i have all the know-how to build such a ting, i am not sure whether the building blocks allow me to do this.
@frank:
Thanks for the suggestion! I will do that ASAP.
Do you published the solution’s source code? Maybe I can help…
Font size…
Gee whats with the small font size on this page, the code blocks are unreadable unless font size is increased by the browser
Since more people have been complaining about the font size/line height i have modified it a bit for both the text and the source code. Thanks for the feedback, i am continously trying to improve the look, so suggestions are welcome!
@Leonardo:
Could you please PM me? I’d prefer to talk this over via e-mail rather than a WordPress comment page 😉
I’ve been working on a detailed project to parse and quantify a complicated course listing website for my college. Unfortunately, the site is a HTML throwback to the early 90’s and does not differentiate between listings in any meaningful way. As a result, the only thing capable of parsing the sea of random tags is a set of carefully constructed regex’s. This is would break very easily if they ever bothered to change how they did their markup, but it works in this case.
As I work on this, I’m constructing a parsing toolkit designed to abstract some of the repetitive regex tasks I frequently go through. While gross overkill for a nicely formatted site, it’s the only thing that seems to work with this html eyesore.
Pingback: /home/chrisdo » WeekBits #25
I would like to talk to you… I have been in a company that commericialized the first two methods you speak of – HTree and REXML. With a GUI designer.
I have a few thoughts about commercial applications… that could be monetized.
@Noel:
You can reach me at peter@[thissite].com. Feel free to send me an email!
Thanks for the techniques listed here.
I’m going to go make a few screens my bitch using these techniques.
Thanks for the mention of rubystuff.com. That site is itself created by scraping content from CafePress, using WWW::Mechanize.
Shamless plug; I wrote about that here: http://neurogami.com/cafe-fetcher/
HTree or HTML::XMLParser
It seems HTML::XMLParser is already included in ruby (in either net/http or mechanize or rexml ?) is already included and does pretty much the same thing as HTree without an extra download. Any reason you prefer HTree?
Pingback: Labnotes » Blog Archive » links for 2006-06-22
@RMX:
Well, the reason for this is very prosaic: I did not know HTML::XMLParser beforehand.
I will chcek it out and see what’s the difference between HTree and XMLParser…
Pingback: greg.rubyfr.net»Blog Archive » [En bref] RWN 12, 18 juin 2006
Pingback: Scraping sites with ruby
Pingback: Anonymous
Pingback: Ruby gets a stylish HTML scraper - scrAPI
I am pretty new to this web scraping stuff…can anyone tell me what are the major business usecases for this scraping? i know this web20 mashup’s does this but any commercial application does this?
tia.
Uday — There are a few different business cases I can think of. A primary one is marketing where you might want to build a contact list for your sales force to call or other sorts of targeting. There are many databases online that contain a lot of useful information.
Other times maybe you are trying to automate a process you have to do often. I saw an author who use a technique like this to track sales. There are other examples like tracking ebay bids on certain items that a power seller might find useful. There are many times where you want to take data from a web page and turn it into structured data for your own purposes.
Very helpfull and interesting article. Wanted to ask your opinion on scrAPI aswell. Looking forward to your next article on this subject.
Hi,
When I copy paste your HTree example it gives error:
undefined method `HTree’ for main:Object (NoMethodError)
on the usage of the HTree class. The following seems to work fine:
require ‘open-uri’
require ‘htree/parse’
require ‘htree/rexml’
require ‘rexml/document’
url = “http://www.google.com/search?q=ruby”
open(url) {
|page| pagecontent = page.read()
doc = HTree.parse(pagecontent).torexml
doc.root.eachelement(‘//a[@class=”l”]’) {
|elem| puts elem.attribute(‘href’).value }
}
Rubyful seems to change utf-8 characters, for instance into %nbsp Is this standard behaviour?
Duh sorry about that, I meant to say the is translated into %nbsp
Nice post…
Pingback: ramonsblog » Blog Archive » Rubinrote Wohnungssuche
HTML::XMLParser?
It took me awhile, but I figured out what RMX was talking about. For the curious:
gem install htmltools
then
require_gem ‘htmltools’
require ‘html/xmltree.rb’
parser = HTMLTree::XMLParser.new(false, false)
parser.parsefilenamed(‘my.html’)
doc = parser.document # is a REXML::Document
Check out lib/html/xmltree.rb at http://ruby-htmltools.rubyforge.org/doc/ for more info. Seems to be functionally identical to htree. Slightly easier to install, but also in my very limited testing almost twice as slow.
-chuck
Thanks for the information, I needed a pick me up.
Hey ‘RMX’: I don’t see any HTML::XMLParser in the standard distribution. You would think before sending people on wild goose chases looking in three different places you say it might be (one of which, Mechanize, isn’t even there either) you would be a little more sure of it yourself. Check your facts next time.
@Mark:
Don’t worry about RMX’s tips 😉 There is a better (by far) solution already: HPricot by why. I am working on my Ruby web-extraction framework right now – using HPricot – and I can tell you, it is absolutely the way to go. It is waaaay faster then any other tool, and they say it has also better shaky-html-parsing capabilities. Well, so far I did not have any problems with any page, and it is really, really lightning fast compared to HTRee + REXML or RubyfulSoup.
Ok Peter I’ll check that out. I’ll also look for your web abstraction framework.
yes, I agree with Bob. rubyfulsoup seems to translate html entity references like ” ” and “é” into “%nbsp” and “%eacute” respectively. Of course, the problem might be in the SGML parser code that rubyful soup uses. It sure would be nice if the community could discuss this problem and its solutions in more detail.
Cs
Pingback: Elliott’s blog » Blog Archive » scRUBYt - Hot, New Ruby Web-Scraping Toolkit Released
Instead of altering the gem, you could just add this at the top of your example :
class WWW::Mechanize::Link
def class_name
node.attributes[‘class’]
end
end
Yeah, that’s absolutely true and it’s definitely the Ruby way – unfortunately when I wrote this article I was totally new to Ruby and (coming from Java) I forgot about the possibility to reopen a class…
you could have a look on http://www.knowlesys.com, they provide web data extraction service.
Pingback: Various tools for screen scrapping « Ruby on Rails Development on Windows
Pingback: scRUBYt! » Ruby Web Scraping Tool Guide - a Simple to Learn and Use, yet Powerful Web Scraping Toolkit Written in Ruby