When I was building my first CLI application, I became very familiar with the Nokogiri gem. I don't consider myself an expert, but I am curious about how gem works. This post walks through my process to scrape data using Nokogiri.
Nokogiri is a Ruby gem designed to scrape data from websites. To use it, all you need is a little CSS knowledge, Ruby fundamentals, and patience.
For this walkthrough, I’m using Yoga Journal’s Poses by Type.
This is one of the pages I used for my CLI application. Scraping sites can fail if their content or structure has changed, so this content is accurate as of publication on June 12, 2018. I'll also be using the Google Chrome browser on a Mac. Feel free to follow along!
1. Add Nokogiri, Open-URI, and Pry to your Ruby file
First, install the gems by typing into the terminal:
gem install nokogiri gem install pry
Once installed, require them at the top of your ruby file:
require ‘nokogiri’ require ‘open-uri’ require ‘pry’
Note: Open-URI is not a gem, but a built-in Ruby wrapper that allows HTML documents to be read as a file:. I’m not sure how it relates to the current Ruby series.
2. Save the site you’d like to scrape to a variable
doc = Nokogiri::HTML(open(‘https://www.yogajournal.com/poses/types'))
3. Add a
binding.pry below the variable and run your file.
If you’re following along, the document should look something like this:
require ‘nokogiri’ require ‘open-uri’ require ‘pry’ doc = Nokogiri::HTML(open(‘https://www.yogajournal.com/poses/types')) binding.pry
And your terminal command to run the file would be:
4. Inspect content to scrape on your webpage
Let’s hop over to the website. For part of my application, I'm hoping to create a class of objects where each instance is a type of yoga pose. I’d like to use Nokogiri to scrape all of the pose type names and their associated URLs.
Right click on the page and hit Inspect. Go to the top left corner of the inspect window, and click on the arrow pointing to the middle of the square (the info balloon should say something like, “Select an element in the page to inspect it”).
Hover over an example of the item you’re looking for. This will highlight the corresponding CSS in the Elements window while highlighting the element’s bounds (usually through a colored box). Click on the item to select its CSS in the Elements window.
When hovering, I like to ask myself:
- Is the content I’m looking for stored within this item? Can I see the text, link, image, etc. wrapped within the CSS in the Elements panel?
Yes! between the > < signs, I see “Arm Balance Yoga Poses”.
- What is the CSS selector or HTML tag for this item? Some common examples are
h2. In Chrome, this will be the text that shows up first, in pink.
For this example, the selector is
- Is there any ID or Class information that follows the selector/tag?
Here, the text is a little long. It looks like the class is called
m-card--header-text ng-isolate-scope. Double click on the class information in the Elements panel and copy it.
Tip: in Google Chrome, while hovering your cursor over the item you’d like to scrape, the format already matches what you need to put into your Nokogiri scrape argument! If it’s an id, the text will have a '#' in front of it. If it’s a class, the text will have a '.' in front of it. In Chrome’s representation, id’s are orange and classes are blue.
- Does similar content on the page share CSS selectors/classes, etc? Hover over similar items to see if they share selector/class/id information.
Yes: "Balancing Yoga Poses" has the same CSS as "Arm Balances"!
Let's take this information back to our code.
5. Play around in Pry
The standard structure for scraping using CSS in Nokogiri is
variable.css(‘selector#id.class’), so in this example,
doc.css(‘h2.m-card--header-text ng-isolate-scope’) is the first thing I’ll run.
Hmmm, looks like we’re getting back an empty array. I was a little skeptical about the class because of space in-between 'text' and 'ng'. When using Nokogiri, spaces can indicate moving through different nested levels. I’ve also noticed that successful Nokogiri scrapes are often simple and do not contain the full CSS. Let’s try pairing it down to include only the class name before the space.
Tip: Hit just the up arrow while in the terminal to get the previous typed value.
Now we’re getting somewhere! In the returned data, I can see the strings I want to pull, but I need to add a bit more specificity to get just the pose type category name.
Hit ‘q’ to exit the returned value and get back to a space to type new commands.
To specify the content you’re searching for in Nokogiri, chain additional methods after
#.css. I most frequently use:
#.text —> will pull strings #.attribute('href').value —> will pull urls #.attribute('img').value —> will pull image urls
We’re looking for a string, so let’s try adding
pry(main)> doc.css('h2.m-card--header-text').text => "Arm Balance Yoga PosesBalancing Yoga PosesBinding Yoga PosesChest-Opening Yoga PosesCore Yoga PosesForward Bend Yoga Po sesHip-Opening Yoga PosesInversion Yoga PosesPranayama Exercises & PosesRestorative Yoga PosesSeated Yoga PosesStanding Yog a PosesStrengthening Yoga PosesTwist Yoga PosesYoga BackbendsYoga BandhaYoga Mudras8 Top Yoga Teachers Give Their Best Advi ce for Finding Bravery in Inversions3 Ways to Modify PaschimottanasanaMaster Paschimottanasana in 6 StepsThe Bandha Approac h You Haven't Tried—That Could Change EverythingThe Muscle That Can Make or Break Healthy Shoulders in InversionsMaster Cla ss: How Does Pranayama Help Digestion?This Energy-Boosting Breathwork Is Better Than CaffeineMaster Class: Rodney Yee’s 3-S tep Pranayama Technique for Stillness and PeaceYoga Journal's 2017 Pose of YearRodney Yee's Restorative Yoga Sequence to Pr epare for Pranayama"
We have one long string that has all pose type headlines! To determine if they are accessible as individuals, I like to run the same code, but chain
pry(main)> doc.css('h2.m-card--header-text').first.text => "Arm Balance Yoga Poses”
Awesome. We’ve got you now.
6. Push the content into an array
Return to your web browser. Take a look at the parent of your desired data. Look for an arrow pointing down in the Elements panel.
I tend to have a lot of luck with using only the class for the method's argument when building an array of data scraped from a webpage. Let’s test the following:
Wonderful — our little strings are nestled within this code. Let’s try iterating through it:
doc.css('.m-card--header').collect do |pose_type| pose_type.css('h2').text end
Here, I’m using the
doc variable, saved to the value of the HTML of our desired webpage, and narrow it down to just the data that has a class of
.m-card--header. I’m using the
#collect iterator to go through each child of this class–which in our case, we hope is a pose type category–and push just the text within
h2 tags into an array.
h2? Nokogiri doesn’t require both tags/selectors and classes/ids in every call. Sometimes one or the other will retrieve what you’re looking for. The
.m-card--header class narrows the retrieved content to children of that class. Each headline is encased within an
h2 tag, and the only
h2 tags that are children of
.m-card--header contain the data I’d like to scrape.
=> ["Arm Balance Yoga Poses", "Balancing Yoga Poses", "Binding Yoga Poses", "Chest-Opening Yoga Poses", "Core Yoga Poses", "Forward Bend Yoga Poses", "Hip-Opening Yoga Poses", "Inversion Yoga Poses", "Pranayama Exercises & Poses", "Restorative Yoga Poses", "Seated Yoga Poses", "Standing Yoga Poses", "Strengthening Yoga Poses", "Twist Yoga Poses", "Yoga Backbends", ...
Don’t be afraid to keep playing around until you find what works, or go back to the CSS to see if there may be another way to scrape what you’re looking for.
I noticed some of the headlines for articles at the bottom of the page got pulled as well, which means I’ll have some additional drilling/data verification to do in my CLI application to retrieve just the pose type category names.
Links are a little trickier. To get the link each pose type points to, I wasn’t able to hover, I needed to click on the title and look within the element's data. When I hovered over the 'a' tag above the 'h2' tag in the Elements panel, the element that contained the link on the page highlighted. That helped me confirm the link data I wanted to scrape was encapsulated within the highlighted CSS element.
To return links instead of text, the Nokogiri format is:
And to iterate through them, use just the ‘a’ tag for the collection, followed by the .attribute(‘href’).value chain as the block. To get the links that accompany these pose types, I used:
doc.css('a.m-card—header').collect do |pose_link| pose_link.attribute('href').value end => ["/poses/types/arm-balances", "/poses/types/balancing", "/poses/types/binds", "/poses/types/chest-openers", "/poses/types/core", "/poses/types/forward-bends", "/poses/types/hip-openers", "/poses/types/inversions", "/poses/types/pranayama", "/poses/types/restorative", "/poses/types/seated-twists", "/poses/types/standing", "/poses/types/strength",
And that’s it! The key thing with Nokogiri is to pay close attention to patterns in the CSS for the site you're scraping. Find the patterns, find the data!