script/extract-text.py

import os
import json
from bs4 import BeautifulSoup
import ebooklib
from ebooklib import epub
import re
import xml.etree.ElementTree as ET

folder_path = './books'  # Replace with your folder path
output_file_pattern = './output/output_part_{}.jsonl'  # Pattern for output files
part_size = 1  # Number of files per part
part_counter = 0
file_counter = 0

def correct_french_punctuation(text):
    # Correct spaces before punctuation in French
    text = re.sub(r'\s+([?!:;])', r'\1', text)  # Remove space before punctuation
    text = re.sub(r'([?!:;])\s*', r'\1 ', text)  # Add space after punctuation if not already present
    text = re.sub(r'\s*-\s*', '-', text)
    text = re.sub(r'\s*–\s*', '-', text)
    return text

def find_navpoint_2_in_toc(book):
    toc_item = book.get_item_with_id('ncx')
    if toc_item is None:
        return None
    toc_content = toc_item.get_content()
    namespaces = {'ncx': 'http://www.daisy.org/z3986/2005/ncx/'}
    toc_root = ET.fromstring(toc_content)

    nav_points = toc_root.findall('.//ncx:navPoint', namespaces)
    for nav_point in nav_points:
        if nav_point.attrib.get('id') == 'navpoint-2':
            return nav_point.text if nav_point.text else None
    return None

def find_section_href_in_toc(book, section_title):
    toc_item = book.get_item_with_id('ncx')
    if toc_item is None:
        return None
    toc_content = toc_item.get_content()
    namespaces = {'ncx': 'http://www.daisy.org/z3986/2005/ncx/'}
    toc_root = ET.fromstring(toc_content)
    nav_points = toc_root.findall('.//ncx:navPoint', namespaces)
    for nav_point in nav_points:
        text_elements = nav_point.findall('.//ncx:navLabel/ncx:text', namespaces)
        for text_element in text_elements:
            if text_element.text == section_title:
                content_element = nav_point.find('.//ncx:content', namespaces)
                if content_element is not None:
                    return content_element.attrib['src']
    return None

def extract_content_from_epub(book):
    text = ''
    start_section = find_section_href_in_toc(book, "Avant propos") or find_section_href_in_toc(book, "Premier Chapitre")
    end_section_1 = find_section_href_in_toc(book, "À propos de cette édition électronique")
    end_section_2 = find_section_href_in_toc(book, "Bibliographie – Œuvres complètes")

    # Determine the final end section
    if end_section_1 is not None and end_section_2 is not None:
        end_section = end_section_1 if end_section_1 < end_section_2 else end_section_2
    elif end_section_1 is not None:
        end_section = end_section_1
    else:
        end_section = end_section_2

    extracting = start_section is None  # Start extracting if no specific start section

    for item in book.get_items():
        if item.get_type() == ebooklib.ITEM_DOCUMENT:
            item_id = item.get_name()
            if start_section and start_section in item_id:
                extracting = True
            if end_section and end_section in item_id:
                break
            if extracting or not start_section:
                try:
                    soup = BeautifulSoup(item.get_content(), 'html.parser')
                    for p in soup.find_all('p'):  # Process paragraph by paragraph
                        paragraph = p.get_text(separator='\n')
                        paragraph = paragraph.replace(u'\xa0', ' ')
                        paragraph = correct_french_punctuation(paragraph)
                        text += paragraph + '\n'
                        # Check for end phrases after each paragraph
                        if "FIN" in paragraph:
                            text = text.split("FIN", 1)[0]
                            print("End of book reached")
                            return text
                        elif "la Bibliothèque électronique du Québec" in paragraph:
                            text = text.split("la Bibliothèque électronique du Québec", 1)[0]
                            print("End of book reached")
                            return text
                        elif "ouvrage est le" in paragraph:
                            text = text.split("ouvrage est le", 1)[0]
                            print("End of book reached")
                            return text
                except Exception as e:
                    print(f"Error processing content: {e}")

    if not text:
        print("Fallback: Adding all text as no specific sections were found.")
        for item in book.get_items():
            if item.get_type() == ebooklib.ITEM_DOCUMENT:
                try:
                    soup = BeautifulSoup(item.get_content(), 'html.parser')
                    text += soup.get_text(separator='\n').replace(u'\xa0', ' ') + '\n'
                except Exception as e:
                    print(f"Error in fallback processing: {e}")

    return text




def extract_metadata_from_epub(book):
    metadata = {}
    try:
        metadata['title'] = book.get_metadata('DC', 'title')
        metadata['author'] = book.get_metadata('DC', 'creator')
        metadata['publisher'] = book.get_metadata('DC', 'publisher')
        # Add more metadata fields if needed
    except Exception as e:
        print(f"Error extracting metadata: {e}")
    return metadata

for file in os.listdir(folder_path):
    if file.endswith('.epub'):
        if file_counter % part_size == 0:
            if 'jsonl_file' in locals():
                jsonl_file.close()
            part_counter += 1
            jsonl_file = open(output_file_pattern.format(part_counter), 'w', encoding='utf-8')

        full_path = os.path.join(folder_path, file)
        try:
            book = epub.read_epub(full_path)
            text = extract_content_from_epub(book)
            meta = extract_metadata_from_epub(book)
            jsonl_file.write(json.dumps({"text": text, "meta": meta}, ensure_ascii=False) + '\n')
            file_counter += 1
            print(f"reading file {file}")
        except Exception as e:
            print(f"Error reading file {file}: {e}")

if 'jsonl_file' in locals():
    jsonl_file.close()