README.md

metadata

license: cc-by-nc-4.0
language:
  - uk
  - en
  - ru
pretty_name: Ukrainska Pravda articles in ukr/rus/eng published on or after 01.01.2022
multilinguality:
  - multilingual

This dataset contains the articles from Ukrainska Pravda of the years 2022-2023, in all translations.

The dataset was created as part of the Eval-UA-tion 1.0 benchmark for evaluating Ukrainian language models (github, paper), where the UP-Titles tasks (shamotskyi/up_titles_masked and anilev6/up_titles_unmasked use this dataset.

Basics

One row of the dataset contains an article, title/author/tags in up to three languages (ukr-rus-eng) w/ the corresponding title, author and tags. Different translations of the same article often have inconsistent tags, so the main tags column contains the representations of the tags from all languages (each tag is named after its URI on the UP website).

The mapping of each tag to its URIs and names in all the languages it's present in is fuond in the tags_mapping.json file, found in the metadata. The list of URIs for all downloaded articles can be found there as well.

Files

• Two versions:
  • The version 0.0.1 (split name incomplete) covers articles from 01.01.2022 until 12.12.2023, kept for now as it's used in some other datasets
  • The version 0.0.2 (split name train) is the one you need and contains all articles from 01.01.2022 till 31.12.2023
• File structure:
  • data/train is the full 2y 0.0.2 dataset, the one you need
  • data/incomplete is the old 0.0.1 version
  • metadata/ contains the tags mappings and list of downloaded URIs for both versions

The rest

• https://serhii.net/dtb/2023-12-13-231213-1710-ukrainska-pravda-dataset/ is the draft of the relevant thesis section https://github.com/pchr8/eval-UA-tion/ has the paper and the Thesis!
• pchr8/up_crawler is the crawler I wrote to gather this dataset
  
  

For any questions get in touch! me@serhii.net