Hugging Face's logo

Datasets:
desearch
/
subnet-22

Modalities:
Text
Formats:
parquet
Languages:
English
Size:
1M - 10M
Tags:
web-crawl
domains
bittensor
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
subnet-22 / README.md
mirianokradze's picture
mirianokradze
domains: 4,885,074 domains
c85846f verified
preview code
|
Raw History Blame Contribute Delete
3.14 kB
metadata 
license: other
license_name: derived-from-public-lists
language:
  - en
pretty_name: Desearch domains
tags:
  - web-crawl
  - domains
  - bittensor
size_categories:
  - 1M<n<10M
configs:
  - config_name: domains
    data_files: domains/domains.parquet

Desearch domains

The domains in scope for crawling by Desearch, Bittensor subnet 22.

Contents

domains/domains.parquet

column type description
host string Registrable domain (eTLD+1), lowercase, without scheme or www.

Sources

The union of six public domain rankings, each entry resolved to its registrable domain using the Public Suffix List:

Filtering

Removed from the union:

  • Country-code TLDs outside English-speaking markets
  • Adult, gambling, malware, phishing, cryptojacking, stalkerware, warez, hacking and DDoS domains
  • Banking portals, URL shorteners, redirectors, advertising and tracking endpoints, dynamic DNS, DNS-over-HTTPS resolvers and residential proxies
  • Social networks, forums, chat, webmail and file hosting, where pages are generated per user rather than published
  • CDNs, certificate authorities, registrars and other infrastructure hostnames

Classification uses the UT1 blacklists from Université Toulouse 1 Capitole together with public adult-domain blocklists.

Every surviving domain is then checked twice more:

  • It must resolve. A domain with no address record is dropped. Rankings are built from historical traffic, so a meaningful share of any of them has since lapsed.
  • It must serve under its own name. A domain that redirects to a different registrable domain is dropped in favour of its destination. Listing both would send two crawlers to one server under two names, and would give the same pages to two different miners.

Inclusion means a domain passed all of these. It does not mean the domain has been crawled. robots.txt is read and obeyed at request time, and a domain that disallows the DesearchBot token is never fetched. Domains that later prove unreachable, disallow DesearchBot, or publish in a language other than English are dropped when the list is next updated.

Crawler

Desearch identifies as DesearchBot and signs its requests with Web Bot Auth, so operators can cryptographically verify the traffic. Requests to a host are sent no more than one per second, and less often where robots.txt specifies a longer Crawl-delay.

To request removal, see desearch.ai/crawler.

License

Derived from the public sources listed above, each of which retains its own terms.