Extracción de información de archivos PDF con Python: una guía completa

Este script demuestra un poderoso flujo de trabajo para procesar archivos PDF, extraer texto, tokenizar oraciones y realizar modelado de temas con visualización, diseñado para un análisis eficiente y profundo.

Descripción general de las bibliotecas

• os: Proporciona funciones para interactuar con el sistema operativo.
• matplotlib.pyplot: se utiliza para crear visualizaciones estáticas, animadas e interactivas en Python.
• nltk: Natural Language Toolkit, un conjunto de bibliotecas y programas para el procesamiento del lenguaje natural.
• pandas: Biblioteca de análisis y manipulación de datos.
• pdftotext: Biblioteca para convertir documentos PDF a texto sin formato.
• re: proporciona operaciones de coincidencia de expresiones regulares.
• seaborn: Biblioteca de visualización de datos estadísticos basada en matplotlib.
• nltk.tokenize.sent_tokenize: Función NLTK para tokenizar una cadena en oraciones.
• top2vec: Biblioteca para modelado de temas y búsqueda semántica.
• wordcloud: Biblioteca para crear nubes de palabras a partir de datos de texto.

Configuración inicial

Importar módulos

import os
import matplotlib.pyplot as plt
import nltk
import pandas as pd
import pdftotext
import re
import seaborn as sns
from nltk.tokenize import sent_tokenize
from top2vec import Top2Vec
from wordcloud import WordCloud
from cleantext import clean

A continuación, asegúrese de que el tokenizador punkt esté descargado:

nltk.download('punkt')

Normalización de texto

def normalize_text(text):
    """Normalize text by removing special characters and extra spaces,
    and applying various other cleaning options."""

    # Apply the clean function with specified parameters
    cleaned_text = clean(
        text,
        fix_unicode=True,  # fix various unicode errors
        to_ascii=True,  # transliterate to closest ASCII representation
        lower=True,  # lowercase text
        no_line_breaks=False,  # fully strip line breaks as opposed to only normalizing them
        no_urls=True,  # replace all URLs with a special token
        no_emails=True,  # replace all email addresses with a special token
        no_phone_numbers=True,  # replace all phone numbers with a special token
        no_numbers=True,  # replace all numbers with a special token
        no_digits=True,  # replace all digits with a special token
        no_currency_symbols=True,  # replace all currency symbols with a special token
        no_punct=False,  # remove punctuations
        lang="en",  # set to 'de' for German special handling
    )

    # Further clean the text by removing any remaining special characters except word characters, whitespace, and periods/commas
    cleaned_text = re.sub(r"[^\w\s.,]", "", cleaned_text)
    # Replace multiple whitespace characters with a single space and strip leading/trailing spaces
    cleaned_text = re.sub(r"\s ", " ", cleaned_text).strip()

    return cleaned_text

Extracción de texto PDF

def extract_text_from_pdf(pdf_path):
    with open(pdf_path, "rb") as f:
        pdf = pdftotext.PDF(f)
    all_text = "\n\n".join(pdf)
    return normalize_text(all_text)

Tokenización de oraciones

def split_into_sentences(text):
    return sent_tokenize(text)

Procesamiento de múltiples archivos

def process_files(file_paths):
    authors, titles, all_sentences = [], [], []
    for file_path in file_paths:
        file_name = os.path.basename(file_path)
        parts = file_name.split(" - ", 2)
        if len(parts) != 3 or not file_name.endswith(".pdf"):
            print(f"Skipping file with incorrect format: {file_name}")
            continue

        year, author, title = parts
        author, title = author.strip(), title.replace(".pdf", "").strip()

        try:
            text = extract_text_from_pdf(file_path)
        except Exception as e:
            print(f"Error extracting text from {file_name}: {e}")
            continue

        sentences = split_into_sentences(text)
        authors.append(author)
        titles.append(title)
        all_sentences.extend(sentences)
        print(f"Number of sentences for {file_name}: {len(sentences)}")

    return authors, titles, all_sentences

Guardar datos en CSV

def save_data_to_csv(authors, titles, file_paths, output_file):
    texts = []
    for fp in file_paths:
        try:
            text = extract_text_from_pdf(fp)
            sentences = split_into_sentences(text)
            texts.append(" ".join(sentences))
        except Exception as e:
            print(f"Error processing file {fp}: {e}")
            texts.append("")

    data = pd.DataFrame({
        "Author": authors,
        "Title": titles,
        "Text": texts
    })
    data.to_csv(output_file, index=False, quoting=1, encoding='utf-8')
    print(f"Data has been written to {output_file}")

Cargando palabras vacías

def load_stopwords(filepath):
    with open(filepath, "r") as f:
        stopwords = f.read().splitlines()
    additional_stopwords = ["able", "according", "act", "actually", "after", "again", "age", "agree", "al", "all", "already", "also", "am", "among", "an", "and", "another", "any", "appropriate", "are", "argue", "as", "at", "avoid", "based", "basic", "basis", "be", "been", "begin", "best", "book", "both", "build", "but", "by", "call", "can", "cant", "case", "cases", "claim", "claims", "class", "clear", "clearly", "cope", "could", "course", "data", "de", "deal", "dec", "did", "do", "doesnt", "done", "dont", "each", "early", "ed", "either", "end", "etc", "even", "ever", "every", "far", "feel", "few", "field", "find", "first", "follow", "follows", "for", "found", "free", "fri", "fully", "get", "had", "hand", "has", "have", "he", "help", "her", "here", "him", "his", "how", "however", "httpsabout", "ibid", "if", "im", "in", "is", "it", "its", "jstor", "june", "large", "lead", "least", "less", "like", "long", "look", "man", "many", "may", "me", "money", "more", "most", "move", "moves", "my", "neither", "net", "never", "new", "no", "nor", "not", "notes", "notion", "now", "of", "on", "once", "one", "ones", "only", "open", "or", "order", "orgterms", "other", "our", "out", "own", "paper", "past", "place", "plan", "play", "point", "pp", "precisely", "press", "put", "rather", "real", "require", "right", "risk", "role", "said", "same", "says", "search", "second", "see", "seem", "seems", "seen", "sees", "set", "shall", "she", "should", "show", "shows", "since", "so", "step", "strange", "style", "such", "suggests", "talk", "tell", "tells", "term", "terms", "than", "that", "the", "their", "them", "then", "there", "therefore", "these", "they", "this", "those", "three", "thus", "to", "todes", "together", "too", "tradition", "trans", "true", "try", "trying", "turn", "turns", "two", "up", "us", "use", "used", "uses", "using", "very", "view", "vol", "was", "way", "ways", "we", "web", "well", "were", "what", "when", "whether", "which", "who", "why", "with", "within", "works", "would", "years", "york", "you", "your", "suggests", "without"]
    stopwords.extend(additional_stopwords)
    return set(stopwords)

Filtrar palabras vacías de temas

def filter_stopwords_from_topics(topic_words, stopwords):
    filtered_topics = []
    for words in topic_words:
        filtered_topics.append([word for word in words if word.lower() not in stopwords])
    return filtered_topics

Generación de nube de palabras

def generate_wordcloud(topic_words, topic_num, palette='inferno'):
    colors = sns.color_palette(palette, n_colors=256).as_hex()
    def color_func(word, font_size, position, orientation, random_state=None, **kwargs):
        return colors[random_state.randint(0, len(colors) - 1)]

    wordcloud = WordCloud(width=800, height=400, background_color='black', color_func=color_func).generate(' '.join(topic_words))
    plt.figure(figsize=(10, 5))
    plt.imshow(wordcloud, interpolation='bilinear')
    plt.axis('off')
    plt.title(f'Topic {topic_num} Word Cloud')
    plt.show()

Ejecución principal

file_paths = [f"/home/roomal/Desktop/Dreyfus-Project/Dreyfus/{fname}" for fname in os.listdir("/home/roomal/Desktop/Dreyfus-Project/Dreyfus/") if fname.endswith(".pdf")]

authors, titles, all_sentences = process_files(file_paths)

output_file = "/home/roomal/Desktop/Dreyfus-Project/Dreyfus_Papers.csv"
save_data_to_csv(authors, titles, file_paths, output_file)

stopwords_filepath = "/home/roomal/Documents/Lists/stopwords.txt"
stopwords = load_stopwords(stopwords_filepath)

try:
    topic_model = Top2Vec(
        all_sentences,
        embedding_model="distiluse-base-multilingual-cased",
        speed="deep-learn",
        workers=6
    )
    print("Top2Vec model created successfully.")
except ValueError as e:
    print(f"Error initializing Top2Vec: {e}")
except Exception as e:
    print(f"Unexpected error: {e}")

num_topics = topic_model.get_num_topics()
topic_words, word_scores, topic_nums = topic_model.get_topics(num_topics)
filtered_topic_words = filter_stopwords_from_topics(topic_words, stopwords)

for i, words in enumerate(filtered_topic_words):
    print(f"Topic {i}: {', '.join(words)}")

keywords = ["heidegger"]
topic_words, word_scores, topic_scores, topic_nums = topic_model.search_topics(keywords=keywords, num_topics=num_topics)
filtered

_search_topic_words = filter_stopwords_from_topics(topic_words, stopwords)

for i, words in enumerate(filtered_search_topic_words):
    generate_wordcloud(words, topic_nums[i])

for i in range(reduced_num_topics):
    topic_words = topic_model.topic_words_reduced[i]
    filtered_words = [word for word in topic_words if word.lower() not in stopwords]
    print(f"Reduced Topic {i}: {', '.join(filtered_words)}")
    generate_wordcloud(filtered_words, i)

Reducir el número de temas.

reduced_num_topics = 5
topic_mapping = topic_model.hierarchical_topic_reduction(num_topics=reduced_num_topics)

# Print reduced topics and generate word clouds
for i in range(reduced_num_topics):
    topic_words = topic_model.topic_words_reduced[i]
    filtered_words = [word for word in topic_words if word.lower() not in stopwords]
    print(f"Reduced Topic {i}: {', '.join(filtered_words)}")
    generate_wordcloud(filtered_words, i)