manestay/world_state_media_multi

World State Media Text (Multilingual, FineWeb-2)

Dataset Description

manestay/world_state_media_multi is the URL index manestay/world_state_media_urls_multi with the FineWeb-2 document text attached: the same 38,755,459 rows, in the same order, plus id, dump, language_script, text and date copied from FineWeb-2. Each row is a FineWeb-2 document whose URL host belongs to a state-run media outlet listed by State Media Monitor (via manestay/state_run_media).

• Documents: 38,755,459
• Language configurations: 274 (one config per FineWeb-2 language/script, plus default)
• Size: ~44 GB (300 Parquet files, zstd)

Schema

Column	Type	Description
url	string	Document URL (as in FineWeb-2)
id	string	FineWeb-2 document id (<urn:uuid:...>)
dump	string	Common Crawl dump the document came from (FineWeb-2 dump)
language	string	FineWeb-2 language configuration, e.g. deu_Latn
language_script	string	Script from FineWeb-2, e.g. Latn
row_index	int64	0-based row position within that configuration's train split in FineWeb-2
base_url	string	Matched outlet host
country	string	Country/territory of the outlet (State Media Monitor)
typology_2022 … typology_2025	string	State Media Monitor typology for each year (raw values)
text	string	Document text from FineWeb-2
date	string	Crawl date from FineWeb-2

State Media Monitor Typology Codes

• SC: State-Controlled Media
• CaPu: Captured Public / State-Managed Media
• CaPr: Captured Private Media
• ISFM: Independent State-Funded & State-Managed Media
• ISF: Independent State-Funded Media
• ISM: Independent State-Managed Media
• IP: Independent Public Media

Some outlets have other raw values (e.g. a note instead of a code) or no label for a given year; these are kept as published.

Source and Matching

• Text source: HuggingFaceFW/fineweb-2, train split, commit af9c13333eb981300149d5ca60a8e9d659b276b9. Every row was checked to have the same url as the FineWeb-2 row at (language, row_index), and a FineWeb-2 language/script equal to its configuration.
• Outlet list: State Media Monitor (640 media companies, 180 countries and territories).
• Matching: exact equality between the document's host (lowercased, one leading www. removed) and the outlet hosts. Numbered variants such as www1. are listed as separate hosts where relevant; other subdomains are not matched.

Usage

from datasets import load_dataset

ds = load_dataset("manestay/world_state_media_multi", "deu_Latn", split="train")
print(ds[0]["base_url"], ds[0]["country"], ds[0]["typology_2025"])
print(ds[0]["text"][:300])

Counts by Language Configuration

Total documents: 38,755,459 across 274 configurations.

Configuration	Documents	Configuration	Documents
rus_Cyrl	7,105,241	und_Hebr	302
arb_Arab	3,309,706	olo_Latn	295
tur_Latn	3,231,246	nya_Latn	264
fra_Latn	2,929,026	sma_Latn	258
spa_Latn	2,317,956	fuv_Latn	237
pol_Latn	2,258,925	cnh_Latn	226
deu_Latn	2,028,777	sms_Latn	224
swe_Latn	1,557,763	dik_Latn	208
hun_Latn	1,543,005	rup_Latn	194
cmn_Hani	1,397,738	kiu_Latn	192
vie_Latn	1,247,118	lat_Latn	184
bos_Latn	1,017,567	ceb_Latn	171
nob_Latn	599,639	tuk_Arab	170
fas_Arab	550,511	cos_Latn	137
fin_Latn	542,676	tel_Telu	117
cat_Latn	526,568	tly_Latn	108
ron_Latn	435,973	apc_Arab	106
ukr_Cyrl	410,495	tsn_Latn	105
por_Latn	403,626	ayr_Latn	96
ces_Latn	403,375	crh_Latn	94
nld_Latn	357,400	snd_Arab	93
ita_Latn	330,464	hyw_Armn	83
slv_Latn	322,557	nds_Latn	80
bul_Cyrl	289,746	und_Cyrl	75
lit_Latn	246,741	yor_Latn	73
dan_Latn	237,588	frp_Latn	72
srp_Cyrl	210,416	lin_Latn	63
jpn_Jpan	206,218	szl_Latn	62
azj_Latn	179,147	tat_Cyrl	60
kaz_Cyrl	177,169	fkv_Latn	59
bel_Cyrl	162,818	hil_Latn	57
zsm_Latn	155,780	ibo_Latn	55
kat_Geor	120,719	xho_Latn	52
khm_Khmr	108,180	lmo_Latn	50
nno_Latn	93,107	luo_Latn	47
als_Latn	90,874	sid_Latn	43
ind_Latn	82,957	sco_Latn	42
urd_Arab	80,004	abk_Cyrl	38
ekk_Latn	77,567	ffm_Latn	37
kor_Hang	74,165	crs_Latn	35
hrv_Latn	72,902	gug_Latn	35
ell_Grek	68,268	und_Mymr	34
khk_Cyrl	65,174	dyu_Latn	33
mkd_Cyrl	61,903	pis_Latn	33
tha_Thai	57,000	bar_Latn	27
swh_Latn	56,777	pcd_Latn	26
isl_Latn	54,403	wuu_Hani	25
roh_Latn	45,680	lld_Latn	24
srp_Latn	41,153	tah_Latn	23
hau_Latn	39,791	toi_Latn	23
arz_Arab	38,163	kam_Latn	19
hin_Deva	38,115	nde_Latn	19
pbt_Arab	37,539	fro_Latn	18
tam_Taml	32,807	und_Grek	18
cym_Latn	31,393	und_Thai	17
uig_Arab	29,479	rcf_Latn	16
amh_Ethi	27,680	hin_Latn	15
uzn_Cyrl	25,016	qup_Latn	13
slk_Latn	24,620	ydd_Hebr	13
eus_Latn	23,495	ksw_Mymr	12
mlt_Latn	22,078	hac_Arab	11
mya_Mymr	21,916	bak_Cyrl	9
ary_Arab	21,661	fij_Latn	8
kir_Cyrl	20,844	lim_Latn	8
som_Latn	20,578	rhg_Latn	8
tuk_Latn	19,885	war_Latn	8
uzn_Latn	19,341	zgh_Tfng	8
npi_Deva	19,207	frr_Latn	7
heb_Hebr	18,604	gcf_Latn	7
lug_Latn	14,995	kaa_Latn	7
ltz_Latn	14,443	lue_Latn	7
sin_Sinh	14,422	mer_Latn	7
gla_Latn	14,218	quy_Latn	7
uig_Latn	12,731	ibg_Latn	6
mri_Latn	11,407	mar_Deva	6
lvs_Latn	11,085	szy_Latn	6
sme_Latn	10,903	wln_Latn	6
ben_Beng	10,770	arg_Latn	5
hye_Armn	10,300	krj_Latn	5
gle_Latn	10,259	kwy_Latn	5
uig_Cyrl	8,893	mhr_Cyrl	5
oss_Cyrl	8,033	ton_Latn	5
tat_Latn	7,845	acm_Arab	4
ckb_Arab	7,646	ilo_Latn	4
bod_Tibt	7,114	inh_Cyrl	4
rmy_Latn	6,890	kan_Knda	4
lao_Laoo	6,519	kea_Latn	4
kmr_Latn	6,260	mck_Latn	4
ndo_Latn	6,183	nap_Latn	4
div_Thaa	6,092	pag_Latn	4
azb_Arab	6,010	pfl_Latn	4
zul_Latn	5,775	aii_Syrc	3
uzs_Arab	5,524	brh_Arab	3
tgk_Cyrl	4,205	cor_Latn	3
sna_Latn	3,629	fao_Latn	3
kin_Latn	2,730	fur_Latn	3
hsb_Latn	2,529	hat_Latn	3
epo_Latn	2,527	kik_Latn	3
gaz_Latn	2,243	ltg_Latn	3
kua_Latn	2,179	sot_Latn	3
fil_Latn	2,050	twx_Latn	3
bis_Latn	1,689	vls_Latn	3
nrm_Latn	1,340	ayp_Arab	2
sdh_Arab	1,308	bew_Latn	2
run_Latn	1,247	cjk_Latn	2
kab_Latn	1,162	diq_Latn	2
gsw_Latn	1,156	ext_Latn	2
pan_Guru	1,090	fub_Latn	2
mal_Mlym	1,084	glk_Arab	2
bam_Latn	1,064	guz_Latn	2
tir_Ethi	1,051	kmb_Latn	2
afr_Latn	1,024	pnb_Arab	2
ars_Arab	1,007	thk_Latn	2
glg_Latn	902	yom_Latn	2
tpi_Latn	896	anp_Deva	1
tzm_Tfng	799	bxr_Cyrl	1
guj_Gujr	782	cni_Latn	1
fit_Latn	765	dsb_Latn	1
bre_Latn	758	fry_Latn	1
gag_Latn	710	fuh_Latn	1
loz_Latn	709	gcr_Latn	1
oci_Latn	672	ike_Cans	1
yue_Hani	623	lad_Latn	1
her_Latn	553	lun_Latn	1
fuf_Latn	481	mas_Latn	1
ast_Latn	465	ndc_Latn	1
kwn_Latn	444	nso_Latn	1
dzo_Tibt	436	rmc_Latn	1
und_Hira	410	rml_Latn	1
und_Kana	389	sah_Cyrl	1
rmn_Latn	385	srd_Latn	1
smo_Latn	381	syl_Latn	1
naq_Latn	362	tso_Latn	1
mww_Latn	343	udm_Cyrl	1
smn_Latn	336	umb_Latn	1
smj_Latn	320	vro_Latn	1
aeb_Arab	313	wls_Latn	1

Notes

• Exact-host matching misses outlet content on hosts that are not in the outlet list (e.g. unlisted subdomains), and includes everything published on a listed host.
• Typology labels describe the outlet, not individual documents.
• FineWeb-2 has no per-dump partitions; documents are grouped by language only.
• Only FineWeb-2's train splits were used; its test splits and *_removed configurations (documents filtered out by FineWeb-2) are not included.
• Duplicates across dumps. FineWeb deduplicates documents within each Common Crawl dump rather than across dumps (FineWeb blog post), while FineWeb-2 deduplicates across dumps within each language. The World State Media datasets inherit this, so the same article can appear more than once, mostly in the English data. Deduplicating at the corpus level would leave about 58M articles instead of 118M (English 79M → 21M, multilingual 39M → 37M). We keep every document because these datasets are meant as a lens into LLM training corpora, and 118M is what those corpora contain.

License and Terms of Use

Released under the Open Data Commons Attribution License (ODC-By) v1.0, the same license as FineWeb-2. The text comes from Common Crawl, so use is also subject to the Common Crawl Terms of Use. The documents themselves remain the property of their publishers.

Contact

bryanli.ca@gmail.com

Attribution & Citation

If you use this dataset, please cite the underlying sources.

State Media Monitor

@misc{statemediamonitor,
  title={State Media Monitor},
  author={{Media and Journalism Research Center}},
  year={2026},
  url={https://statemediamonitor.com}
}

FineWeb-2

@misc{penedo2025fineweb2pipelinescale,
  title={FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language},
  author={Guilherme Penedo and Hynek Kydlíček and Vinko Sabolčec and Bettina Messmer and Negar Foroutan and Amir Hossein Kargaran and Colin Raffel and Martin Jaggi and Leandro Von Werra and Thomas Wolf},
  year={2025},
  eprint={2506.20920},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2506.20920},
}