BramVanroy/CommonCrawl-CreativeCommons

The Common Crawl Creative Commons Corpus (C5)

Raw CommonCrawl crawls, annotated with Creative Commons license information

C5 is an effort to collect Creative Commons-licensed web data in one place.

The licensing information is extracted from the web pages based on whether they link to Creative Commons licenses either overtly in a tags (like in the footer of Wikipedia) or in metadata fields indicating deliberate Creative Commons publication. However, false positives may occur! See Recommendations and Caveats below! Also see Personal and Sensitive Information.

Code

I am very grateful to the Flemish Supercomputer to provide compute necessary to create this dataset, but as you can tell there is still a lot of data left to be processed. Therefore, I am happy to collaborate to process as many Common Crawl crawls as possible. Shoot me a message if you want to sponsor this project with compute! You can also simply run the code yourself if you'd like. You can find the whole code base, based on datatrove, on Github. If you use the code, please reference my work accordingly and share your processed crawls with the rest of the world (or get in touch with me so I can add them to this repo).

The approach to creating this dataset is different from similar endeavors such as the awesome common-pile/dolma-cccc and C4Corpus datasets. They rely on intricately crafted regular expressions to quickly extract potential licenses from a web page (string-based matching). However, doing so makes it hard to retrieve any structural meta information about the license such as where it was found on the page. In C5, the whole webpage is parsed into a programmatic structure, allowing for an iterative search through this parsed "tree". That makes it possible to track where licenses were found (in the head of a document, for instance). Such information is crucial to minimise false positives: if a license is referred in a meta tag in the head of an HTML page, it is more trustworthy than a "random link" referring to a copyright license in the middle of a web page, which might just be discussing the license in general or providing a license for a picture on the website. Metadata about the license is powerful to attach confidence to the extracted licenses, enabling robust filtering to avoid false positives. While I strongly believe this approach is valuable it also makes it very slow compared to a regex search!

Usage

from datasets import load_dataset

# Everything, most recent -- massive, you will need streaming
ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", streaming=True)

# v1 (2019-30, 2020-05, 2022-05, 2023-06, 2024-51, 2025-05, 2024-46)
ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "v1", streaming=True)

# Single dump, all languages -- large, you may need streaming on non-server hardware
ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "CC-MAIN-2019-30")

# Single language, all dumps -- very large, you will likely need streaming
ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "nld", streaming=True)

# Single language, single dump
ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "CC-MAIN-2019-30-nld")

Progress

In the v1 release, the following crawls are included

• CC-MAIN-2019-30
• CC-MAIN-2020-05
• CC-MAIN-2023-06
• CC-MAIN-2024-51
• CC-MAIN-2024-46
• CC-MAIN-2025-05
• CC-MAIN-2022-05

Languages

The following languages are included. This is a limited set due to computational and storage limitations.

• Afrikaans: afr
• German: deu
• English: eng
• French: fra
• Frysian: fry
• Italian: ita
• Dutch: nld
• Spanish: spa

Quantity

Detailed number of tokens (Llama 3.3 tokenizer) and number of documents are given in the counts.json file.

Language	Number of Documents	Number of Tokens
afr	312,262	358,873,448
deu	9,530,746	11,362,859,534
eng	92,635,372	87,537,859,958
fra	9,234,900	12,366,480,025
fry	230,910	197,430,774
ita	10,734,597	11,913,669,333
nld	2,827,636	2,757,074,705
spa	22,226,944	22,515,709,432
Total	147,733,367	149,009,957,209

Fields

In some cases, multiple licenses are found on a single page. All licenses are collected in potential_licenses. These are then sorted based on three criteria (first option is most preferred, last option is least preferred, e.g. a license found in a meta tag is more trustworthy than a license in an a tag, a license in a footer is more trustworthy than a license not in the footer of a page).

1. location_preference_order: meta_tag, json-ld, link_tag, a_tag
2. head_preference_order: True, False
3. footer_preference_order: True, False

Based on these criteria, the "best" license is picked as the one in the license_* columns. Potential disagreement between multiple licenses is given in license_disagreement.

• text: the extracted text (unmodified)
• id: WARC-Record-ID
• dump: Common Crawl crawl
• url: original url for document
• date: crawl date
• file_path: file path on the S3 bucket
• license_abbr: the license type. Possible values: "cc-unknown" (recommended to filter this one out), "by", "by-sa", "by-nd", "by-nc", "by-nc-sa", "by-nc-nd", "zero", "certification", "mark". If multiple licenses were found (potential_licenses)
• license_version: the license version, e.g. "4.0"
• license_location: the location where the license was found. Possible values: "meta_tag", "json-ld", "link_tag", "a_tag"
• license_in_head: whether the license was found inside a head HTML element
• license_in_footer: whether the license was found inside a footer HTML element, or an HTML element that had footer in the ID or class name
• potential_licenses:
  • abbr: list of all found license abbreviations
  • version: list of all found license versions
  • location: list of all found license locations
  • in_head: list of whether licenses were found in the head
  • in_footer: list of whether licenses were found in a footer
• license_parse_error: whether there was a problem when trying to extract the license, e.g. an unparseable HTML document
• license_disagreement: whether the potential_licenses["abbr"] disagree, i.e., different types of licenses were found. License versions are not included in the comparison!
• language: the language, as detected by glotlid
• language_score: the language identification confidence score
• found_in_fw: whether this sample was found in FineWeb(-2). For non-English, crawls that are more recent than FW2 (everything after 2024-18) is marked as None. For English, crawls that are more recent than FW v1.3 is marked as None (after 2024-51).

Recommendations and Caveats

• Raw CommonCrawl data is processed in an attempt to extract licensing information. No quality filtering is done!! It is highly recommended to filter this data further on quality, fluency, toxicity, etc.
• Similarly, the data has not been deduplicated.
• The licenses include all possible Creative Commons licenses, including non-commercial ones. Take care about what kind of data you wish to use, and filter out non-commercial licenses when needed.
• The column license_disagreement indicates whether multiple licenses were found that have not the same abbreviation, e.g. cc-by and cc-by-nc. It is recommended to filter these out.
• The column license_parse_error indicates whether an error occurred when parsing the license. You probably want to filter out documents where this was the case, though this should be extremely rare.
• Unsurpisingly, the data contains a lot of Wikipedia/Wikimedia content. Depending on what you need, you may wish to filter those out. For Wikipedia specifically, you may opt to use the more thoroughly parsed (but potentially more outdated) wikimedia/wikipedia set.
• In exceptional cases, a link to creativecommons.org is found but the exact license could not be found. These are under license_abbr="cc-unknown" which you may wish to filter out.

Based on these recommendations, two subsets are available:

• BramVanroy/CommonCrawl-CreativeCommons-fine: only retaining items containing samples that are also in FineWeb(-2)
• BramVanroy/CommonCrawl-CreativeCommons-strict: additional filtering that also removes Wikipedia and non-commercial data

Personal and Sensitive Information

C5 is a heavily filtered version of the Common Crawl dataset. CommonCrawl respects robots.txt and will not include websites if their robots.txt say so. Even so, if you find that your website was included you can submit a removal request indicating that you are the owner of the website.

Take-down notices on other Common Crawl-based datasets such as FineWeb are considered. Domains specified and verified in those take-down notices are not included in this dataset.

In this dataset, measures are taken to anonymise email addresses and public IP addresses following the FineWeb-2 approach. Email addresses matching a regular expression are replaced with firstname.lastname@example.org. Similarly, IP addresses allocated for public networks are replaced by unused IP addresses. Despite these best efforts on such large volumes of text, you may still encounter that your personal information is present in the dataset. In that case you can submit a removal request.

Citation

In the current absence of a publication, please cite the dataset as follows. Including a footnote url to this page is also appreciated!

@misc{vanroy2025C5,
    author       = { Bram Vanroy },
    title        = { CommonCrawl CreativeCommons Corpus (C5) },
    year         = 2025,
    url          = { https://huggingface.co/datasets/BramVanroy/CommonCrawl-CreativeCommons },
    doi          = { 10.57967/hf/5340 },
    publisher    = { Hugging Face }
}

If you use or modify the software, please cite:

@software{Vanroy_CommonCrawl-CreativeCommons_2025,
  author = {Vanroy, Bram},
  license = {GPL-3.0},
  month = feb,
  title = {{CommonCrawl-CreativeCommons}},
  url = {https://github.com/BramVanroy/CommonCrawl-CreativeCommons},
  version = {1.3.0},
  year = {2025}
}

Acknowledgments

• The Common Crawl non-profit organization.
• The computational resources and services used in this work were provided by the VSC (Flemish Supercomputer Center), funded by the Research Foundation Flanders (FWO) and the Flemish Government – department EWI under grant 2024-107.
• Guilherme Penedo (@guipenedo) and the rest of the FineWeb and datatrove team for the help and insights
• TNO, who funded the work hours to accomplish this code. They intend to use (parts of) the generated material for the GPT-NL project.
• ML6 and specifically Robin Van Craenenbroek for their Fondant Creative Commons filter for image datasets. While my approach is different, their code did serve as inspiration.