ksolovev/fine-news-sample

Fine-News Sample

Fine-News Sample contains 1,000,000 rows sampled from the Fine-News corpus. The sample covers all 117 capture months and 388 language-and-script labels in that corpus. Each selected row preserves its article text, source metadata, and sampling weight. The full corpus derives from INFINI-NEWS, which extracts article text from Common Crawl News web archives.

At a glance

Measure	Value
Rows	1,000,000
Distinct document IDs	1,000,000
Sum of source sampling weights	1,703,976
Capture months	117 (2016-08 through 2026-04)
Language-and-script labels	388
Parquet files	117
Compressed data size	5.60 GB
Source corpus rows	635,068,343

Use revision="v1" to load Fine-News Sample v1.

Coverage

Capture months record when the crawler retrieved a page. The article's publication date comes from page metadata when available. Language labels combine an ISO 639-3 language code with an ISO 15924 script code. The counts describe classifier assignments; classification accuracy requires separate evidence.

Capture year	Sample rows
2016	5,998
2017	55,959
2018	71,350
2019	93,758
2020	139,151
2021	139,648
2022	137,684
2023	130,504
2024	106,087
2025	96,544
2026	23,317

Sampling method

Sampling uses each nonempty capture-month and language partition as a stratum, meaning a group sampled separately. Each stratum receives one row to preserve coverage of rare labels. The remaining rows are allocated in proportion to the remaining stratum populations, using the largest-remainder method. Within each stratum, the recipe selects row positions uniformly without replacement from the sorted source Parquet files. The recipe uses seed 0, SHA256-derived stratum seeds, and Python 3.12 random.sample. sampling.json records each stratum's population and sample size, the source commit, and sample file hashes. The recipe is available at research/sample_recipe.py.

Rare strata receive a larger sampling fraction because each stratum receives at least one row. For population totals, use inverse inclusion weights: each stratum's population divided by its sample size. The preserved metadata.sampling_weight values describe the full corpus's duplicate-family weighting. Sample selection and source weighting require separate treatment in an analysis.

Schema

Files use the path data/<capture-month>/part-000000.parquet. Each monthly file contains the selected language groups. The top-level columns match Fine-News:

Column	Type	Description
text	string	Filtered article text.
id	string	Source warc_record_id, retained as the document identifier.
media	list of structs	Optional DataTrove media field.
metadata	struct	Source fields, language evidence, duplicate-family fields, and sampling weight.

The metadata struct preserves source URLs, Web ARChive (WARC) provenance, capture timestamps, publication metadata, and extraction versions. Use metadata.capture_month for capture month, metadata.language for the assigned language-and-script label, and metadata.sampling_weight for source training weights.

Quick start

The default Hugging Face view loads text, id, and metadata. Use PyArrow to read all four Parquet columns, including media.

from datasets import load_dataset

rows = load_dataset(
    "ksolovev/fine-news-sample",
    revision="v2",
    split="train",
    streaming=True,
)
for row in rows.take(3):
    print(row["id"], row["metadata"]["language"])

The train view contains all sample rows. Load one capture year with name="year_2024"; year subsets cover 2016 through 2026. Users define any training and test splits required for their work.

Limitations

Coverage, language classification, and filtering vary across publishers, languages, and time. Stratified sampling preserves month and language coverage; the sample's representativeness requires separate evidence. Article text can contain personal information, errors, and harmful content. The Fine-News card describes the source corpus's filtering and duplicate removal.

Terms of use

Article text remains subject to its publishers' copyright and source-specific terms. Fine-News assigns no dataset-wide text license. Users must assess the rights and terms that apply to their intended use, including the INFINI-NEWS terms. Those terms restrict bulk article text to academic research. INFINI-NEWS releases its derived metadata under CC-BY-4.0.

Citation

If you use the sample, cite Fine-News Sample and the INFINI-NEWS paper:

@misc{solovev2026finenewssample,
  author = {Solovev, Kirill},
  title = {{Fine-News Sample}},
  year = {2026},
  version = {2},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/ksolovev/fine-news-sample}
}