wikipedia_fr / extract_wiki /3_extract_txt.py
eckendoerffer's picture
Upload 6 files
8ae2c07
# -*- coding: utf-8 -*-
"""
Wikipedia Text Extractor:
This script consists of two main stages, transforming raw HTML source code to testing and filtering lines to retain.
Required:
pip install beautifulsoup4 langid
Step 1: /html_wiki/ -> /txt_wiki/
- Removal of undesired tags and classes.
- Extraction of text content from the HTML code.
Step 2: /txt_wiki/ -> /txt_wiki_lines/
- Sentence-by-sentence testing to determine elements for exclusion based on criteria like count of special characters, language detection, etc.
- Language testing is performed on each line and also for every element within parentheses.
The two stages have been deliberately separated. Additionally, numerous log files are generated to aid in debugging, testing, and refining exclusion parameters.
The goal here is to have a fruitful harvest for some quality text juice, not to hit the lottery jackpot numbers.
Author : Guillaume Eckendoerffer
Date : 22-09-23
Repository : https://github.com/Eckendoerffer/TorchTrainerFlow/
https://huggingface.co/datasets/eckendoerffer/wikipedia_fr
"""
import os, re
from bs4 import BeautifulSoup
from langid.langid import LanguageIdentifier, model
identifier = LanguageIdentifier.from_modelstring(model, norm_probs=True)
MAX_LENGTH = int(330) # nb words per line
MAX_SENTENCE_LENGTH = int(350) # max words per sentence
# LANG EXCLUDE
ALLOWED_LANGUAGE = ['fr']
# MATH EXCLUDE
MAX_NUM_COUNT = 26
MAX_PLUS_SIGN_COUNT = 5
MAX_EQUALS_SIGN_COUNT = 5
# CHAR EXCLUDE
MAX_DOUBLE_QUOTE_COUNT = 18
MAX_PARENTHESIS_COUNT = 14
MAX_BRACKET_COUNT = 12
MAX_COMMA_COUNT = 40
MAX_DOLLAR_COUNT = 5
# LONG EXCLUDE
LONG_WORD_CHARS = 29 # LONG_WORD size in CHARS
MAXIMUM_LONG_WORDS = 3 # Remove line if LONG_WORDS count > MAXIMUM_LONG_WORDS
# CLASS EXCLUDE
CLASSES_TO_REMOVE = ['bandeau-container', 'bandeau-section', 'metadata', 'bandeau-niveau-information', 'gallery', 'infobox_v3']
# TAG EXCLUDE
TAG_TO_REMOVE = ['nav', 'menu', 'ul', 'ol', 'table', 'h1', 'h2', 'h3', 'h4', 'h5']
# PATH
PATH = os.path.dirname(os.path.abspath(__file__)).replace('\\', '/')
HTML_PATH = PATH + '/sources/html_wiki/'
TARGET_LONG = PATH + "/excluded_long.txt"
TARGET_LANG = PATH + "/excluded_lang.txt"
TARGET_MATH = PATH + "/excluded_math.txt"
TARGET_CHARS = PATH + "/excluded_chars.txt"
TARGET_PARENTH = PATH + "/excluded_parentheses.txt"
TARGET_LATEX = PATH + "/excluded_latex.txt"
FILES = [f for f in os.listdir(HTML_PATH) if os.path.isfile(os.path.join(HTML_PATH, f))]
with open(TARGET_LONG, 'w', encoding="utf8") as f:
f.write("")
with open(TARGET_LANG, 'w', encoding="utf8") as f:
f.write("")
with open(TARGET_MATH, 'w', encoding="utf8") as f:
f.write("")
with open(TARGET_CHARS, 'w', encoding="utf8") as f:
f.write("")
with open(TARGET_PARENTH, 'w', encoding="utf8") as f:
f.write("")
with open(TARGET_LATEX, 'w', encoding="utf8") as f:
f.write("")
def extract_wikipedia_text(html_path, txt_path):
""" Step 1: Extraction of text content from the HTML code """
with open(html_path, "r", encoding="utf-8") as f:
wiki_content = f.read()
soup = BeautifulSoup(wiki_content, 'html.parser')
# Remove menus, sub-menus, lists, and tables
for tag in soup.find_all(TAG_TO_REMOVE):
tag.decompose()
# Find and remove the divs from classes to remove
for class_name in CLASSES_TO_REMOVE:
for div in soup.find_all("div", class_=class_name):
div.decompose()
# Retrieve only the text
text = soup.get_text()
text = text.replace(chr(8217), "'")
text = text.replace("`", "'")
text = text.replace("‘", "'")
text = re.sub(r'\[\d+\]', ' ', text)
text = re.sub(r'\{[^\}]*\}', ' ', text)
text = re.sub(r'\[[^\}]*\]', ' ', text)
with open(txt_path, "w", encoding="utf-8") as f:
f.write(text)
return len(text)
def split_into_sentences(text):
sentences = re.split(r'([.;!?]\s*\u00BB|[.;!?\u00BB]\s*(?!\u00BB)|\s*--\s*)', text)
sentences = ["".join(i) for i in zip(sentences[0::2], sentences[1::2])]
return sentences
def text_standardize(text):
text = text.replace('—', '-')
text = text.replace('–', '-')
text = text.replace('―', '-')
text = text.replace('…', '...')
text = re.sub('''(~+|!+|"+|;+|\?+|\++|,+|\)+|\(+|\\+|\/+|\*+|\[+|\]+|}+|{+|\|+|_+)''', r' \1 ', text)
text = re.sub('\s*\n\s*', ' \n ', text)
text = re.sub('[^\S\n]+', ' ', text)
text = re.sub(r"\s{2,}", " ", text)
return text.strip()
def has_repeated_uniq_char(text):
pattern = r'([a-zA-Z0-9] ){5,}'
return bool(re.search(pattern, text))
def countLongText(text):
nb_long = 0
t = text.split()
for w in t:
if len(w) > LONG_WORD_CHARS:
if w.count('-')!= 0:
nb_long += 1
return nb_long
def remove_latex(text):
sp_chars = 0
sp_chars += text.count('(')
sp_chars += text.count(')')
sp_chars += text.count('{')
sp_chars += text.count('}')
sp_chars += text.count('_')
sp_chars += text.count('/')
sp_chars += text.count(' ')
if sp_chars > (len(text)/2) and len(text) > 4:
text = ''
return text
def extract_parentheses(text):
stack = []
results = []
for i, char in enumerate(text):
if char == '(':
stack.append(i)
elif char == ')' and stack:
start = stack.pop()
results.append((start, i))
return results
def is_date_or_year_range(content):
return bool(re.match(r'^\d{4}(-\d{4})?$', content.strip()))
def remove_language_in_parentheses(line, target_file_parentheses):
for start, end in reversed(extract_parentheses(line)):
match = line[start+1:end]
if is_date_or_year_range(match):
continue
lang = identifier.classify(match)
if lang[0] not in ALLOWED_LANGUAGE:
line = line[:start] + line[end+1:]
target_file_parentheses.write(f'({match})' + "\n")
return line
def test_exclude(line_add, target_file_math, target_file_chars, target_file_lang, target_file_long, target_file, target_file_parentheses):
nb_words_line = len(line_add.split())
if countLongText(line_add.strip()) > MAXIMUM_LONG_WORDS \
or len( re.findall( r'\d+ ', line_add.strip() ) ) > MAX_NUM_COUNT \
or line_add.count('=') > MAX_EQUALS_SIGN_COUNT or line_add.count('+') > MAX_PLUS_SIGN_COUNT:
target_file_math.write(f"{line_add.strip()} \n")
elif line_add.count('"') > MAX_DOUBLE_QUOTE_COUNT or line_add.count('(') > MAX_PARENTHESIS_COUNT \
or line_add.count('[') > MAX_BRACKET_COUNT \
or line_add.count(',') > MAX_COMMA_COUNT \
or line_add.count('$') > MAX_DOLLAR_COUNT:
target_file_chars.write(f"{line_add.strip()} \n")
else:
lang = identifier.classify(line_add)
if lang[0] not in ALLOWED_LANGUAGE:
target_file_lang.write(f"[{lang[0]}] {line_add.strip()} \n")
else:
if len(line_add.split()) > MAX_SENTENCE_LENGTH or has_repeated_uniq_char(line_add.strip()):
target_file_long.write(f"[{nb_words_line}] {line_add.strip()} \n")
else:
line_add = re.sub(r"\s{2,}", " ",remove_language_in_parentheses(line_add.strip(), target_file_parentheses))
target_file.write(f"{line_add} \n")
def test_line(full_path, full_target_path):
""" Step 2: Exclusion based on criteria """
nb_words_line = 0
line_add = ""
with open(full_target_path, 'w', encoding="utf8") as f:
f.write("")
with open(full_path, "r", encoding="utf8", errors="ignore") as source_file, \
open(full_target_path, "a", encoding="utf8") as target_file, \
open(TARGET_LONG, "a", encoding="utf8") as target_file_long, \
open(TARGET_LANG, "a", encoding="utf8") as target_file_lang, \
open(TARGET_MATH, "a", encoding="utf8") as target_file_math, \
open(TARGET_CHARS, "a", encoding="utf8") as target_file_chars, \
open(TARGET_PARENTH, "a", encoding="utf8") as target_file_parentheses, \
open(TARGET_LATEX, "a", encoding="utf8") as target_file_latex:
for line in source_file:
line = '' if line.count('Articles détaillés :') else line.strip()
line = line.replace("Un article de Wikipédia, l'encyclopédie libre.", "")
if line.count('Ce document provient de') \
or line.count('https://') \
or line.count('wikipedia.') \
or line.count('index.') \
or line.count('php?') \
or line.count('title='):
line=''
line = re.sub(r',\s,', ' ', line)
line = re.sub(r'\.\s\.', '.', line)
line = re.sub(r',\s\.', '.', line)
sentences = split_into_sentences(re.sub(r"\s{2,}", " ", line))
for sentence in sentences:
if remove_latex(text_standardize(sentence)) =='':
target_file_latex.write(f"{sentence.strip()} \n")
sentence =''
words = len(sentence.split())
if len(line_add.split()) + words < MAX_LENGTH:
nb_words_line += words
line_add += f" {text_standardize(sentence)}"
else:
test_exclude(line_add, target_file_math, target_file_chars, target_file_lang, target_file_long, target_file, target_file_parentheses)
nb_words_line = len(sentence.split())
line_add = f" {text_standardize(sentence)}"
if nb_words_line:
test_exclude(line_add, target_file_math, target_file_chars, target_file_lang, target_file_long, target_file, target_file_parentheses)
for i, file in enumerate(FILES):
html_path = HTML_PATH + file
txt_path = html_path.replace('html_wiki', 'txt_wiki')
txt_len = extract_wikipedia_text(html_path, txt_path)
txt_lines_path = html_path.replace('html_wiki', 'txt_wiki_lines')
test_line(txt_path, txt_lines_path)
print(f"({i+1}/{len(FILES)}) {file} {txt_len}")