Web Scraping and Crawling with Scrapy and MongoDB

Getting Started

There are two possible ways to continue from where we left off.

The first is to extend our existing Spider by extracting every next page link from the response in the parse_item method with an xpath expression and just yield a Request object with a callback to the same parse_item method. This way scrapy will automatically make a new request to the link we specify. You can find more information on this method in the Scrapy documentation.

The other, much simpler option is to utilize a different type of spider - the CrawlSpider(link). It’s an extended version of the basic Spider, designed exactly for our use case.

The CrawlSpider

We’ll be using the same Scrapy project from the last tutorial, so grab the code from the repoif you need it.

Create the Boilerplate

Within the “stack” directory, start by generating the spider boilerplate from the crawltemplate:

$ scrapy genspider stack_crawler stackoverflow.com -t crawl
Created spider 'stack_crawler' using template 'crawl' in module:
  stack.spiders.stack_crawler

The Scrapy project should now look like this:

├── scrapy.cfg
└── stack
    ├── __init__.py
    ├── items.py
    ├── pipelines.py
    ├── settings.py
    └── spiders
        ├── __init__.py
        ├── stack_crawler.py
        └── stack_spider.py

And the stack_crawler.py file should look like this:

# -*- coding: utf-8 -*-
import scrapy
from scrapy.contrib.linkextractors import LinkExtractor
from scrapy.contrib.spiders import CrawlSpider, Rule

from stack.items import StackItem


class StackCrawlerSpider(CrawlSpider):
    name = 'stack_crawler'
    allowed_domains = ['stackoverflow.com']
    start_urls = ['http://www.stackoverflow.com/']

    rules = (
        Rule(LinkExtractor(allow=r'Items/'), callback='parse_item', follow=True),
    )

    def parse_item(self, response):
        i = StackItem()
        #i['domain_id'] = response.xpath('//input[@id="sid"]/@value').extract()
        #i['name'] = response.xpath('//div[@id="name"]').extract()
        #i['description'] = response.xpath('//div[@id="description"]').extract()
        return i

We just need to make a few updates to this boilerplate…

Update the `start_urls` list

First, add the first page of questions to the start_urls list:

start_urls = [
    'http://stackoverflow.com/questions?pagesize=50&sort=newest'
]

Update the `rules` list

Next, we need to tell the spider where it can find the next page links by adding a regular expression to the rules attribute:

rules = [
    Rule(LinkExtractor(allow=r'questions\?page=[0-9]&sort=newest'),
         callback='parse_item', follow=True)
]

Scrapy will now automatically request new pages based on those links and pass the response to the parse_item method to extract the questions and titles.

If you’re paying close attention, this regex limits the crawling to the first 9 pages since for this demo we do not want to scrape all 176,234 pages!

Update the `parse_item` method

Now we just need to write how to parse the pages with xpath, which we already did in the last tutorial - so just copy it over:

def parse_item(self, response):
    questions = response.xpath('//div[@class="summary"]/h3')

    for question in questions:
        item = StackItem()
        item['url'] = question.xpath(
            'a[@class="question-hyperlink"]/@href').extract()[0]
        item['title'] = question.xpath(
            'a[@class="question-hyperlink"]/text()').extract()[0]
        yield item

That’s it for the spider, but do not start it just yet.

Add a Download Delay

We need to be nice to StackOverflow (and any site, for that matter) by setting a download delay in settings.py:

DOWNLOAD_DELAY = 5

This tells Scrapy to wait at least 5 seconds between every new request it makes. You’re essentially rate limiting yourself. If you do not do this, StackOverflow will rate limit you; and if you continue to scrape the site without imposing a rate limit, your IP address could be banned. So, be nice - Treat any site you scrape as if it were your own.

Now there is only one thing left to do - store the data.

MongoDB

Last time we only downloaded 50 questions, but since we are grabbing a lot more data this time, we want to avoid adding duplicate questions to the database. We can do that by using a MongoDB upsert, which means we update the question title if it is already in the database and insert otherwise.

Modify the MongoDBPipeline we defined earlier:

class MongoDBPipeline(object):

    def __init__(self):
        connection = pymongo.MongoClient(
            settings['MONGODB_SERVER'],
            settings['MONGODB_PORT']
        )
        db = connection[settings['MONGODB_DB']]
        self.collection = db[settings['MONGODB_COLLECTION']]

    def process_item(self, item, spider):
        for data in item:
            if not data:
                raise DropItem("Missing data!")
        self.collection.update({'url': item['url']}, dict(item), upsert=True)
        log.msg("Question added to MongoDB database!",
                level=log.DEBUG, spider=spider)
        return item

For simplicity, we did not optimize the query and did not deal with indexes since this is not a production environment.

Test

Start the spider!

$ scrapy crawl stack_crawler

Now sit back and watch your database fill with data!

$ mongo
MongoDB shell version: 3.0.4
> use stackoverflow
switched to db stackoverflow
> db.questions.count()
447
>

TechiCrunch

Search This Blog