README.md

metadata

language:
  - sk
license: cc-by-4.0
size_categories:
  - 10K<n<100K
task_categories:
  - text-classification
tags:
  - clustering
  - slovak
  - news
  - mteb
pretty_name: Pravda.sk URL-based Clustering Dataset
dataset_info:
  features:
    - name: url
      dtype: string
    - name: title
      dtype: string
    - name: summary
      dtype: string
    - name: content
      dtype: string
    - name: date
      dtype: string
    - name: tags
      sequence: string
    - name: url_category
      dtype: string
    - name: url_subdomain
      dtype: string
    - name: url_section
      dtype: string
  splits:
    - name: test
      num_examples: 15000

Pravda.sk URL-based Clustering Dataset

A Slovak news article clustering dataset based on URL structure from pravda.sk, designed for evaluating text clustering and embedding models.

Dataset Description

This dataset contains 15,000 Slovak news articles categorized by their URL structure. Articles are organized into 50 categories across 11 subdomains, providing a hierarchical classification based on editorial decisions.

Use Case

This dataset is designed for:

• Evaluating Slovak text embedding models
• Clustering benchmark tasks (MTEB)
• Text classification experiments
• Slovak NLP research

Dataset Structure

Fields

Field	Type	Description
url	string	Original article URL
title	string	Article headline
summary	string	Article summary/lead paragraph
content	string	Full article text
date	string	Publication date (ISO format)
tags	list[string]	Original pravda.sk tags
url_category	string	Full category path (e.g., sportweb/tenis)
url_subdomain	string	Subdomain (e.g., sportweb)
url_section	string	Section within subdomain (e.g., tenis)

Statistics

• Total articles: 15,000
• Categories: 50
• Samples per category: 300
• Subdomains: 11

Subdomains

Subdomain	Articles	Description
sportweb	2,100	Sports news
vat	2,100	Science & technology
koktail	1,800	Celebrity & entertainment
spravy	1,800	General news
cestovanie	1,500	Travel
ekonomika	1,200	Business & economy
kultura	1,200	Culture & arts
zurnal	1,200	Magazine/features
auto	900	Automotive
zdravie	900	Health
uzitocna	300	Practical/lifestyle

Data Quality

• No duplicates: Deduplicated by title + summary combination
• No null values: All articles have valid title and summary fields
• Balanced: 300 samples per category

Usage

from datasets import load_dataset

dataset = load_dataset("NaiveNeuron/pravda-sk-url-clustering")

# Access the test split
articles = dataset["test"]

# Example: Get articles by category
tennis_articles = [a for a in articles if a["url_category"] == "sportweb/tenis"]

Source

Articles were collected from pravda.sk via web.archive.org snapshots spanning 2004-2025.

License

This dataset is released under CC-BY-4.0. Please check pravda.sk terms of service for content usage rights.

Citation

If you use this dataset, please cite:

@dataset{pravda_sk_url_clustering,
  title = {Pravda.sk URL-based Clustering Dataset},
  author = {NaiveNeuron},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/NaiveNeuron/pravda-sk-url-clustering}
}