credi-net/DomainPool

Dataset Card for domain-pool 0.1.1

domain-pool is a fine grained and cross-domain aggregate labelled set of 15,999,167 web domains.

These web domains have up to 11 features, including 5 grading axes:

• Features:
  • year;
  • website category (e.g. news or adult);
  • country, or if applicable, perpetrator and / or target country (e.g. in the case of coordinated campaigns).
• Scoring axes:
  • reliability (may be a continous, categorical or binary score);
  • factuality (same);
  • bias (may be continous or categorical);
  • popularity (as a rank). All domains also have the original data source indicated per label, along with their year to enable temporal analyses. A large part of these data sources are open-sources academic datasets, as well as sourced from fact-checking organisations, governmental or cybersecurity investigations, and more.

The full composition is provided below.

Dataset Overview

Label composition

Domain Features

These characterize the domain, with:

• Year: of each dataset the domain was present in;
• Type: the broader category the website belongs to (e.g. phishing or adult);
• Country: the domain may have one country associated, or in certain cases (e.g. targeted campaigns), have a perpetrator and/or atarget country.

The prominent types are represented below:

Moreover, all datapoints and labels are timestamped. Most data (in terms of volume) is sourced from recent or regularly updated, as reflected in the 2026 prominence shown below:

Reliability

Reliability as a broad category encompasses three types of quantitative labels at different granularities:

1. Continuous score ( n = ): these are numerical (float) on [0.0,1.0] that explicitly relates to the domain's reliability as assessed by expert fact-checkers (independent or academic).
2. Binary ( n = ): a boolean flag ('(un)reliable') indicating broader reliability.
3. 3-class ( n = ): same type of source and meaning, these span three levels: [low, medium, high].
4. 6-class ( n = ): same, at a finer granularity.

More precisely,

Reliability (continuous)

Distribution:

Reliability (binary)

We have a third 'N/A'-like category, for 'providers', in the sense of domains that are not responsible for their content either becuase they are a social media platform, a media or file hosting service, or another platform of the likes.

We also look at agreement rates between the 10 largest datasets (for readability), where overlaps with fewer than 50 samples are not represented.

Reliability (3-class)

We also look at agreement rates between datasets:

Reliability (6-class)

Factuality (continuous)

Factuality (3-class)

Bias (continuous)

Bias (categorical)

Popularity metrics

The Pool has 3 types of popularity metric: iffy_rank, mbfc_rank and tranco_rank, from the datasets of the same name. Iffy and Tranco are relative ranking, while MBFC has traffic-relative categories:

Data sources

Some of the primary contributors to the dataset are:

• UT1 by the University of Toulouse Capitole (41.5%),
• The Tranco List (28.27%).
• Blacklists (30.8%);
• 50+ others with <10% each.

The full list:

Source	Rows	% of Total
UT1	6,644,316	41.5%
Tranco	4,944,640	30.9%
Blacklists	4,931,489	30.8%
Malicious and Benign Webpages (Train)	1,200,000	7.5%
Malicious URLs	651,191	4.1%
Benign & Malicious URLs	632,508	3.9%
Phish DB	496,442	3.1%
RADEK (Benign C)	436,811	2.7%
Malicious and Benign Webpages (Test)	361,934	2.3%
RADEK (Benign U)	360,708	2.3%
RADEK (Phishing)	164,425	1.0%
DNS Blocklists	142,877	0.89%
URL Phish	103,011	0.64%
RADEK (Malware)	100,809	0.63%
HOSTS (Adware & Malware)	82,622	0.52%
HOSTS (Porn)	76,721	0.48
URLHaus	75,180	0.47%
LegitPhish	63,984	0.40%
Phish Dataset	44,265	0.28%
domains-quality-ratings	11,520	0.07%
Meta Threat Reports	6,379	0.04%
HOSTS (Gambling)	6,027	0.04%
Redirection Domains	5,751	0.04%
MisinfoDomains	4,767	0.0298%
MBFC Ratings	4,497	0.0281%
Wikipedia (General)	3,935	0.0246%
Manual	2,323	0.0145%
HOSTS (Fake News)	2,186	0.0137%
Iffy Index	2,040	0.0128%
MBFC's Questionable List	1,883	0.0118%
FakeNewsNet	1,130	0.0071%
CheckThat!	1,067	0.0067%
Wikipedia (Campaigns)	844	0.0053%
Wikipedia (Miscellaneous)	832	0.0052%
Providers	708	0.0044%
Providers (Manual)	708	0.0044%
Wikipedia (Fake News)	461	0.0029%
Reliable Legal Resources	393	0.0025%
Zoznam	337	0.0021%
Politifact's Almenac	327	0.0020%
Reliable Health Resources	325	0.0020%
Dictionaries (Manual)	304	0.0019%
SD22 Approved Software List	263	0.0016%
Paperwall	123	0.0008%
NGO Report (UAE Blacklist)	100	0.0006%
NGO Report (Israel Blacklist)	99	0.0006%
NGO Report (Saudi Blacklist)	86	0.0005%
Wikipedia (II Actors)	65	0.0004%
Nelež	51	0.0003%
Tools (Manual)	48	0.0003%
C2 Domains	20	0.0001%
Hasbara Tracker	19	0.0001%
EDMO Hubs	16	0.0001%
NGO Report (Russia Blacklist)	4	0.0000%

• Curated by the CrediNet organisation, which consists of a team of collaborators from the Complex Data Lab @ Mila - Quebec AI Institute, the University of Oxford, McGill University, Concordia University, UC Berkeley, University of Montreal, and AITHYRA.
• Funding: This research was supported by the Engineering and Physical Sciences Research Council (EPSRC) and the AI Security Institute (AISI) grant: Towards Trustworthy AI Agents for Information Veracity and the EPSRC Turing AI World-Leading Research Fellowship No. EP/X040062/1 and EPSRC AI Hub No. EP/Y028872/1. This research was also enabled in part by compute resources provided by Mila (mila.quebec) and Compute Canada.

Data sources:

• UT1 by the University of Toulouse Capitole,
• DQR by Lin et al.),
• Wikipedia,
• Lasser et al..