comcrawl

A python utility for downloading Common Crawl data.


Keywords
commoncrawl, machine-learning, nlp, data, common-crawl, scraping, deep-learning, training-dataset, python
License
MIT
Install
pip install comcrawl==1.0.1

Documentation

comcrawl

GitHub Workflow Status codecov GitHub

comcrawl is a python package for easily querying and downloading pages from commoncrawl.org.

Introduction

I was inspired to make comcrawl by reading this article.

Note: I made this for personal projects and for fun. Thus this package is intended for use in small to medium projects, because it is not optimized for handling gigabytes or terrabytes of data. You might want to check out cdx-toolkit or cdx-index-client in such cases.

What is Common Crawl?

The Common Crawl project is an "open repository of web crawl data that can be accessed and analyzed by anyone". It contains billions of web pages and is often used for NLP projects to gather large amounts of text data.

Common Crawl provides a search index, which you can use to search for certain URLs in their crawled data. Each search result contains a link and byte offset to a specific location in their AWS S3 buckets to download the page.

What does comcrawl offer?

comcrawl simplifies this process of searching and downloading from Common Crawl by offering a simple API interface you can use in your python program.

Installation

comcrawl is available on PyPI.

Install it via pip by running the following command from your terminal:

pip install comcrawl

Usage

Basic

The HTML for each page will be available as a string in the 'html' key in each results dictionary after calling the download method.

from comcrawl import IndexClient

client = IndexClient()

client.search("reddit.com/r/MachineLearning/*")
client.download()

first_page_html = client.results[0]["html"]

Multithreading

You can leverage multithreading while searching or downloading by specifying the number of threads you want to use.

Please keep in mind to not overdo this, so you don't put too much stress on the Common Crawl servers (have a look at Code of Conduct).

from comcrawl import IndexClient

client = IndexClient()

client.search("reddit.com/r/MachineLearning/*", threads=4)
client.download(threads=4)

Removing duplicates & Saving

You can easily combine this package with the pandas library, to filter out duplicate results and persist them to disk:

from comcrawl import IndexClient
import pandas as pd

client = IndexClient()
client.search("reddit.com/r/MachineLearning/*")

client.results = (pd.DataFrame(client.results)
                  .sort_values(by="timestamp")
                  .drop_duplicates("urlkey", keep="last")
                  .to_dict("records"))

client.download()

pd.DataFrame(client.results).to_csv("results.csv")

The urlkey alone might not be sufficient here, so you might want to write a function to compute a custom id from the results' properties for the removal of duplicates.

Searching subsets of Indexes

By default, when instantiated, the IndexClient fetches a list of currently available Common Crawl indexes to search. You can also restrict the search to certain Common Crawl Indexes, by specifying them as a list.

from comcrawl import IndexClient

client = IndexClient(["2019-51", "2019-47"])
client.search("reddit.com/r/MachineLearning/*")
client.download()

Logging HTTP requests

When debugging your code, you can enable logging of all HTTP requests that are made.

from comcrawl import IndexClient

client = IndexClient(verbose=True)
client.search("reddit.com/r/MachineLearning/*")
client.download()

Code of Conduct

When accessing Common Crawl, please beware these guidelines posted by one of the Common Crawl maintainers:

https://groups.google.com/forum/#!msg/common-crawl/3QmQjFA_3y4/vTbhGqIBBQAJ