extract_wiki/1_extract_link.py

# -*- coding: utf-8 -*-

"""
Wikipedia URLs Extractor:

Script to download the Wikipedia dataset from Hugging Face, extract URLs,
and save them to a text file for further processing.

Required:
pip install datasets

Using MIN_LENGTH = 1400 results in approximately 1,100,000 URLs.
Using MIN_LENGTH = 1000 results in approximately 1,800,000 URLs.

Author     : Guillaume Eckendoerffer
Date       : 14-09-23
Repository : https://github.com/Eckendoerffer/TorchTrainerFlow/
             https://huggingface.co/datasets/eckendoerffer/wikipedia_fr
"""

import os
from datasets import load_dataset

# Constants
MIN_LENGTH           = 1400  # Minimum text length in number of characters to be added to the URL list.
EXCLUDE_TITLES_START = ['Liste ', 'Abréviations ', '(homonymie)']

# File path configurations
PATH = os.path.dirname(os.path.abspath(__file__))
URL_FILEPATH = os.path.join(PATH, "wiki_link.txt")

# Resetting the output file
with open(URL_FILEPATH, 'w', encoding="utf8") as url_file:
    url_file.write("")

# Loading the dataset
dataset = load_dataset('wikipedia', '20220301.fr')
subset  = dataset["train"]

add = 0
for i, row in enumerate(subset):
    text  = row["text"]
    title = row["title"]
    url   = row["url"]

    # Checking conditions to add the URL
    if not any(title.startswith(e) for e in EXCLUDE_TITLES_START) and len(text) >= MIN_LENGTH:
        add += 1
        print(f"{add} : {len(text)} : {url} : {title}")
        with open(URL_FILEPATH, 'a', encoding="utf8") as url_file:
            url_file.write(f"{url.strip()} \n")