You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Each article and image in this repository is licensed under the licence given for it in this dataset (the license / image_license field and html/manifest.csv). The licences that occur are: Attribution-NonCommercial 4.0 International (https://creativecommons.org/licenses/by-nc/4.0/legalcode), Attribution-NonCommercial-NoDerivatives 4.0 International (https://creativecommons.org/licenses/by-nc-nd/4.0/legalcode), Attribution-NonCommercial-ShareAlike 4.0 International (https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode), Attribution-NoDerivatives 4.0 International (https://creativecommons.org/licenses/by-nd/4.0/legalcode), Attribution-ShareAlike 4.0 International (https://creativecommons.org/licenses/by-sa/4.0/legalcode), Attribution 4.0 International (https://creativecommons.org/licenses/by/4.0/legalcode), and public domain. The full legal texts are in the licenses/ folder.

Log in or Sign Up to review the conditions and access this dataset content.

NDLA-KM (gated)

Access-restricted companion to NeddiAS/NDLA-KM. Read that card first: it describes the benchmark, the schema, the splits and how to reproduce the paper.

Experiment code: https://github.com/neddi-as/NDLA-KM

Contents

path what
data/gated-nc-nd/articles.parquet the 1,567 gated-nc-nd articles (40 CC-BY-NC-4.0, 164 CC-BY-NC-ND-4.0, 1,361 CC-BY-NC-SA-4.0, 2 CC-BY-ND-4.0); same schema as open-core
data/image-text/image_text.parquet title and alt text of the images whose own licence is NC/ND
html/html-0000{0..3}-of-00004.tar the raw page HTML of all 25,597 text-tier articles (open-core and gated-nc-nd), one file per article
html/manifest.csv article id, tier, licence, sha256 and size of every HTML file
html/SHA256SUMS, html/CANARY.txt, html/LEAK-WARNING.md checksums, the reuse marker, and what it is for
licenses/ full legal texts of the CC licences

The 193 test and 9 zero-shot articles in this tier are needed to train and evaluate on the paper's full splits.

Raw HTML and the reuse marker

The raw HTML contains the labels, so it is never used as model input. It is released for corpus analysis. Every file carries the fixed string

NDLA-KM-CANARY:5653377c-1c03-4e68-9347-dcd682b90545

in an HTML comment and in a hidden element whose text survives HTML-to-text extraction, so a future corpus can be probed for it. See html/LEAK-WARNING.md.

hf download NeddiAS/NDLA-KM-gated --repo-type dataset --local-dir data/NDLA-KM-gated
cd data/NDLA-KM-gated/html && for f in html-*.tar; do tar xf "$f"; done && sha256sum -c SHA256SUMS

Articles without a usable licence (unlicensed-idlist) are not in this repo at all.

Downloads last month
-