Datasets:

tshasan
/

CC-Meta25-1M

license: cc-by-4.0
tags:
  - common-crawl
  - web-metadata
  - nlp
  - '2025'
version: 1.0.0
task_categories:
  - text-classification
size_categories:
  - 100K<n<1M

CC-Meta25-1M

Overview

The CC-Meta25-1M dataset comprises 1,000,341 tuples of web metadata (URL, title, snippet, language) extracted via uniform random sampling from the Common Crawl February 2025 dataset (CC-MAIN-2025-08. Released under version 1.0.0, it captures a representative snapshot of web content from early 2025, retaining original snippet text including boilerplate such as cookie notices and navigation elements for authenticity. This dataset is intended for researchers and practitioners exploring web metadata, linguistic distributions, or temporal web trends.

Dataset Specifications

Source: Common Crawl CC-MAIN-2025-08
Size: 1,000,341 records, [323.3 MB] (Parquet format)
Sampling: Uniform random selection from the February 2025 crawl
Features:
- url: Web page URL (string; "N/A" if invalid or missing)
- title: Page title (string; "N/A" if missing)
- snippet: Text excerpt from the page (string; includes boilerplate, "N/A" if missing)
- language: ISO 639-1 language code (string; e.g., "en", "N/A" if missing)
Processing:
- Missing values in all fields replaced with "N/A".
- Duplicates preserved to mirror the raw crawl (0.01% duplicated).

Accessing the Dataset

Hosted on Hugging Face, the dataset can be accessed as follows:

from datasets import load_dataset
dataset = load_dataset("tshasan/cc-meta25-1m")
print(dataset[0])