About the Data Set

Overview

The Common Crawl data set contains approximately 6 billion web documents stored on a publicly accessible, scalable computer cluster.  Here is some more information on the content and storage of the data set.

File Locations

The entire Common Crawl data set is stored on Amazon S3 as a Public Data Set:

  http://aws.amazon.com/datasets/41740

The data set is divided into three major subsets:

  Archived Crawl #1 - s3://aws-publicdatasets/common-crawl/crawl-001/ - crawl data from 2008/2010
  Archived Crawl #2 - s3://aws-publicdatasets/common-crawl/crawl-002/ - crawl data from 2009/2010
  Current Crawl - s3://aws-publicdatasets/common-crawl/parse-output/ - crawl data from 2012

The two archived crawl data sets are stored in folders organized by the year, month, date, and hour the content was crawled.  For example:

  s3://aws-publicdatasets/common-crawl/crawl-002/2010/01/06/10/1262847572760_10.arc.gz

the contents of this file were crawled started on January 6th, 2010 at 10 AM.

The current crawl data set is stored in the "parse-output" folder in a similar manner to how Nutch stores archives.  Crawl data is stored in a "segments" subfolder, then in a folder that starts with the UNIX timestamp of crawl start time.  For example:

  s3://aws-publicdatasets/common-crawl/parse-output/segment/1341690169105/1341826131693_45.arc.gz

The "crawl-intermediate", "merge-output", and "stats-output" subfolders are used for internal data processing, and, while publicly available, are currently not documented and are not considered part of the corpus.

File Types

The current crawl data set includes three different types of files:  ARC raw content, Text Only, and Metadata.

The archived crawl data sets contain only ARC raw content files.

ARC Files - Raw Content

ARC files contain the full HTTP response and payload for all pages crawled. The ARC file format was designed by the Internet Archive.  You can read more about this file format here:

  http://archive.org/web/researcher/ArcFileFormat.php

ARC files are a series of concatenated GZIP documents.  The first compressed member is a ARC file header, which usually looks like this:

filedesc://1341817173109_4.arc.gz 0.0.0.0 20120709065933 text/plain 73
1 0 CommonCrawl
URL IP-address Archive-date Content-type Archive-length

This file header lists the fields that are used in the record header of subsequent records:  URL, IP Address, Archive Date, Content Type, Archive Length

The rest of the individually compressed members consist of an ARC record header, followed by the full HTTP response:

http://www.srlchem.com/products/ 74.55.84.98 20120518232759 text/html 28556
HTTP/1.1 200 OK
Server:nginx
Date:Fri, 18 May 2012 23:28:04 GMT
Content-Type:text/html
...

The ARC files reside in timestamp-based folders in the archived crawls, and in the segment folders in the current crawl.  They files are named "*.arc.gz".  For example:

  s3://aws-publicdatasets/common-crawl/parse-output/segment/1341690169105/1341826131693_45.arc.gz

The Metadata files, described below, contain offsets into the ARC files.  You can use these offsets as a cross-reference between content metadata and the actual content.

Text Files - Text Only

Common Crawl also produces a series of text only files.  These files take content returned as HTML or RSS and parse out just the text content - making it easier for researchers to perform text-based analysis.

Text Only files are saved as Hadoop SequenceFiles using GZIP compression.  The key and value data types are both Text.  The key in these files is the URL, and the value is the actual text content.  From HTML pages, the text content includes the page title, the page meta description content, and all text content from the HTML body.  They are located in the segment directories, with a file name of "textData-nnnnn".  For example:

  s3://aws-publicdatasets/common-crawl/parse-output/segment/1341690169105/textData-00112

The numbers at the end of the filename are sequentially assigned to Text Only files within the segment - they have not specific meaning.

Further, Text Only files are translated from their native character sets into UTF-8.  All Text Only content (in all languages) can be read using the UTF-8 character set.

Currently, we are only producing Text Only files from HTML and RSS/Atom content.  The Text Only files are on average 20% of the size of the raw content.

Metadata

In addition to content files, Common Crawl produces a series of Metadata files that provide useful information about the crawled content.  For each URL, the Metadata files contain status information, the HTTP response code, and file names and offsets of ARC files where the raw content can be found.

Most importantly for some users, the Metadata files contain the HTML title, HTML meta tags, RSS/Atom information, and all anchors/hyperlinks from HTML documents (including all fields on the link tags).  Users can scan the metadata files to pick up extracted links rather than extracting the links themselves.

Records in the Metadata files are in the same order and have the same file numbers as the Text Only content.

Metadata files are saved as Hadoop SequenceFiles using GZIP compression.  The key and value data types are also both Text.  The key in these files is the URL, and the value is a JSON structure of fields and subfields - the full structure is defined below..  Just like the Text Only files, the Metadata files are located in the segment directories, with a file name of "metadata-nnnnn".  For example:

  s3://aws-publicdatasets/common-crawl/parse-output/segment/1341690169105/metadata-00112

Metadata Schema

The Metadata file JSON contains the following attributes.

All attributes are considered optional, and are only present when they apply.

General Attributes

The following attributes apply to all types of content:

HTML Content Attributes

The "content" JSON object of an HTML document can contain the following fields:

RSS Content Attributes

The "content" JSON object of an RSS feed document can contain the following fields:

Atom Content Attributes

The "content" JSON object of an Atom feed document can contain the following fields: