UA-News

Dataset Summary

Ukrainian News is a collection of more than 150 thousand news articles, gathered from more than 20 news resources. Dataset samples are divided into 5 categories: політика, спорт, новини, бізнес, технології. The dataset is provided by the non-profit student's organization FIdo.ai (machine learning research division of FIdo, National University of Kyiv-Mohyla Academy) for research purposes in data mining (classification, clustering, keywords extraction, etc.).

Dataset development is still in progress

Dataset Structure

Parameters:

• root : Directory path

• download: Whether to download data

• split: Which split of the data to load (train or test)

• return_tags: Whether to return text keywords

Splits:

• Train :

  • File size: 324 MB
  • Number of samples: 120417

  • Target distribution

    політика : 40364 (33.5%)

    спорт : 40364 (33.5%)

    новини : 40364 (33.5%)

    бізнес : 40364 (33.5%)

    технології : 40364 (33.5%)

• Test:

  • File size: 81 MB
  • Number of samples: 30105

  • Target distribution

    політика : 40364 (33.5%)

    спорт : 40364 (33.5%)

    новини : 40364 (33.5%)

    бізнес : 40364 (33.5%)

    технології : 40364 (33.5%)

Data sample

{
  "title" : 'На Донеччині зафіксували сьомий випадок коронавірусу',
  "text" : 'Про це повідомив голова Донецької ОДА Павло Кириленко в Facebook ...,
  "tags" : ['Донецька область', 'COVID-19', 'Новини'],
  "target" : 'новини'
 }

Example of usage

Our API

from ua_datasets import NewsClassificationDataset

train_data = NewsClassificationDataset(root='data/', split='train', return_tags=True)

for title, text, tags, target in train_data:
    print(title, text, tags, target)

Hugging Face 🤗 API

from datasets import load_dataset

dataset = load_dataset("FIdo-AI/ua-news")

for item in dataset["train"]:
    title, text, tags, target = item["title"], item["text"], item["tags"], item["target"]
    print("Title: " + title)
    print("Text: " + text)
    print("Tags: " + tags)
    print("Target: " + target)