Web data scraping (online news comments) with Scrapy (Python)

Question

I want to scrape web comments data from online news purely for research. And I noticed that I have to learn about Scrapy...

Usually, I do programming with Python. I though it will be easy to learn. But I got some problems.

I want to scrape news comment in http://news.yahoo.com/congress-wary--but-unlikely-to-blow-up-obama-s-iran-deal-230545228.html.

But the problem is there is a button (>View Comments (452)) to see the comments. In addition, what I want to do is scraping all the comments in that news. Unfortunately, I have to click another button (View more comments) to see other 10 comments more.

How can I handle this problem?

The code that I've done is as below. Sorry for too poor code.

#############################################
from scrapy.spider import BaseSpider
from scrapy.selector import Selector
from tutorial.items import DmozItem

class DmozSpider(BaseSpider):
   name = "dmoz"
   allowed_domains = ["news.yahoo.com"]

   start_urls = ["http://news.yahoo.com/blogs/oddnews/driver-offended-by-%E2%80%9Cwh0-r8x%E2`%80%9D-license-plate-221720503.html",]

   def parse(self, response):
       sel = Selector(response)
       sites = sel.xpath('//div/p')
       items = []
       for site in sites:
           item = DmozItem()
           item['title'] = site.xpath('/text()').extract()
           items.append(item)
       return items  

You can see that how much left to be done to solve my problem. But I have to be hurry.. I will do my best anyway.

Answer 1

Since you seem like the try-first ask-question later type (that's a very good thing), I won't give you an answer, but a (very detailed) guide on how to find the answer.

The thing is, unless you are a yahoo developer, you probably don't have access to the source code you're trying to scrape. That is to say, you don't know exactly how the site is built and how your requests to it as a user are being processed on the server-side. You can, however, investigate the client-side and try to emulate it. I like using Chrome Developer Tools for this, but you can use others such as FF firebug.

So first off we need to figure out what's going on. So the way it works, is you click on the 'show comments' it loads the first ten, then you need to keep clicking for the next ten comments each time. Notice, however, that all this clicking isn't taking you to a different link, but lively fetches the comments, which is a very neat UI but for our case requires a bit more work. I can tell two things right away:

1. They're using javascript to load the comments (because I'm staying on the same page).
2. They load them dynamically with AJAX calls each time you click (meaning instead of loading the comments with the page and just showing them to you, with each click it does another request to the database).

Now let's right-click and inspect element on that button. It's actually just a simple span with text:

<span>View Comments (2077)</span>

By looking at that we still don't know how that's generated or what it does when clicked. Fine. Now, keeping the devtools window open, let's click on it. This opened up the first ten. But in fact, a request was being made for us to fetch them. A request that chrome devtools recorded. We look in the network tab of the devtools and see a lot of confusing data. Wait, here's one that makes sense:

http://news.yahoo.com/_xhr/contentcomments/get_comments/?content_id=42f7f6e0-7bae-33d3-aa1d-3dfc7fb5cdfc&_device=full&count=10&sortBy=highestRated&isNext=true&offset=20&pageNumber=2&_media.modules.content_comments.switches._enable_view_others=1&_media.modules.content_comments.switches._enable_mutecommenter=1&enable_collapsed_comment=1

See? _xhr and then get_comments. That makes a lot of sense. Going to that link in the browser gave me a JSON object (looks like a python dictionary) containing all the ten comments which that request fetched. Now that's the request you need to emulate, because that's the one that gives you what you want. First let's translate this to some normal reqest that a human can read:

go to this url: http://news.yahoo.com/_xhr/contentcomments/get_comments/
include these parameters: {'_device': 'full',
          '_media.modules.content_comments.switches._enable_mutecommenter': '1',
          '_media.modules.content_comments.switches._enable_view_others': '1',
          'content_id': '42f7f6e0-7bae-33d3-aa1d-3dfc7fb5cdfc',
          'count': '10',
          'enable_collapsed_comment': '1',
          'isNext': 'true',
          'offset': '20',
          'pageNumber': '2',
          'sortBy': 'highestRated'}

Now it's just a matter of trial-and-error. However, a few things to note here:

1. Obviously the count is what decides how many comments you're getting. I tried changing it to 100 to see what happens and got a bad request. And it was nice enough to tell me why - "Offset should be multiple of total rows". So now we understand how to use offset

2. The content_id is probably something that identifies the article you are reading. Meaning you need to fetch that from the original page somehow. Try digging around a little, you'll find it.

3. Also, you obviously don't want to fetch 10 comments at a time, so it's probably a good idea to find a way to fetch the number of total comments somehow (either find out how the page gets it, or just fetch it from within the article itself)

4. Using the devtools you have access to all client-side scripts. So by digging you can find that that link to /get_comments/ is kept within a javascript object named YUI. You can then try to understand how it is making the request, and try to emulate that (though you can probably figure it out yourself)

5. You might need to overcome some security measures. For example, you might need a session-key from the original article before you can access the comments. This is used to prevent direct access to some parts of the sites. I won't trouble you with the details, because it doesn't seem like a problem in this case, but you do need to be aware of it in case it shows up.

6. Finally, you'll have to parse the JSON object (python has excellent built-in tools for that) and then parse the html comments you are getting (for which you might want to check out BeautifulSoup).

As you can see, this will require some work, but despite all I've written, it's not an extremely complicated task either.

So don't panic.

It's just a matter of digging and digging until you find gold (also, having some basic WEB knowledge doesn't hurt). Then, if you face a roadblock and really can't go any further, come back here to SO, and ask again. Someone will help you.

Good luck!