Request access to website_crawl_0924
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This dataset contains web pages and images collected from public websites. All content remains the property of its respective owners and is provided for research purposes only. Requests are reviewed manually.
Log in or Sign Up to review the conditions and access this dataset content.
website_crawl_0924
Pages and images of 61 catalog websites (40 topic categories), collected live from
the sites' public origins. Candidate URLs are the sites' complete Common Crawl URL
inventories (all 128 collections, legacy index_all selection) plus links
discovered on saved pages.
Every saved page has:
raw/<key>.origin.html.gz— the HTML exactly as served (gzip).pages/<key>.html.gz— a passive, cleaned HTML rendition (no scripts, styles or forms; ordered media), with the page's images localized once they were downloaded.documents/<key>.json.gz— the manifest: URL, capture time, HTTP metadata, provenance (Common Crawl record or discovering page), quality label, title, ordered media list with each image's local path and SHA-256.
Images are stored once by content under _media/blobs/<sha256>.<ext> and linked
from <category>/<site>/commoncrawl/media/<asset_id>.<ext>. A page belonging to
several catalog sites is stored once and linked into every member site. Per-block
Common Crawl index records are under <site>/commoncrawl/indexes/, raw index
blocks under _meta/cc-index/, and per-URL result logs under _meta/results/.
Layout
The dataset is published in parts while the crawl runs (each part is uploaded, then removed from the crawler's storage):
parts.json every sealed part: files, members, bytes
data/part-NNNNN/*.tar tar shards (pax); member names are the tree paths above
data/part-NNNNN/logs-*.tar result and decision logs of that period
data/part-NNNNN/index.jsonl.gz one line per member: tar file, path, offset, size (or link target)
data/part-NNNNN/manifest.json files with size and SHA-256, member count
tools/unpack.py restore the tree from all parts
tools/verify.py check sizes/SHA-256, index bounds and that no path repeats
meta/sites.json the site catalog
No duplicates, nothing left out. Every path of the tree occurs in exactly one part. Pages are published only in their final rendered version; content fetched again (the same image under another URL) is not published twice. Downloading all parts and unpacking them gives the complete tree, in any order.
Use
hf download l1997i/website_crawl_0924 --repo-type dataset --local-dir website_crawl_0924
python3 website_crawl_0924/tools/verify.py website_crawl_0924
python3 website_crawl_0924/tools/unpack.py website_crawl_0924 --dest tree
Parts can also be downloaded one at a time
(--include "data/part-00003/*"), and single members read straight from a tar
using the offsets in index.jsonl.gz (the tars are standard; tar -xf works too,
except for hard links pointing into another part).
Rights
The pages and images belong to their respective owners. They are shared for research on web content; redistribution of the content itself may require the owners' permission. Access is granted manually.
- Downloads last month
- 461