app.py

import os
import re
import pandas as pd
from PyPDF2 import PdfReader
from transformers import pipeline, AutoTokenizer
from gradio import Interface, File
import gradio as gr
import spaces

# Initialize a list to store the data
data = []

# Load the LED tokenizer and model
led_tokenizer = AutoTokenizer.from_pretrained("allenai/led-base-16384-multi_lexsum-source-long")
classifier = pipeline("text-classification", model="allenai/led-base-16384-multi_lexsum-source-long", tokenizer=led_tokenizer, framework="pt")

# Load the summarization model and tokenizer
summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6", tokenizer="sshleifer/distilbart-cnn-12-6", framework="pt")

# Function to clean text by keeping only alphanumeric characters and spaces
def clean_text(text):
    return re.sub(r'[^a-zA-Z0-9\s]', '', text)

# Function to split text into chunks of a specified size
def split_text(text, chunk_size=1024):
    words = text.split()
    for i in range(0, len(words), chunk_size):
        yield ' '.join(words[i:i + chunk_size])

# Function to classify text using LED model
@spaces.GPU(duration=120)
def classify_text(text):
    try:
        return classifier(text)[0]['label']
    except IndexError:
        return "Unable to classify"

# Function to summarize text using the summarizer model
@spaces.GPU(duration=120)
def summarize_text(text, max_length=100, min_length=30):
    try:
        return summarizer(text, max_length=max_length, min_length=min_length, do_sample=False)[0]['summary_text']
    except IndexError:
        return "Unable to summarize"

# Function to extract a title-like summary from the beginning of the text
@spaces.GPU(duration=120)
def extract_title(text, max_length=20):
    try:
        return summarizer(text, max_length=max_length, min_length=5, do_sample=False)[0]['summary_text']
    except IndexError:
        return "Unable to extract title"

# Define the folder path and CSV file path
# output_folder_path = '/content/drive/My Drive/path_to_output'  # Adjust this to your actual path

# Define the Gradio interface for file upload and download
@spaces.GPU(duration=120)
def process_files(pdf_files):
    for pdf_file in pdf_files:
        text = extract_text(pdf_file)

        # Skip encrypted files
        if text is None:
            continue

        # Extract a title from the beginning of the text
        title_text = ' '.join(text.split()[:512])  # Take the first 512 tokens for title extraction
        title = extract_title(title_text)

        # Initialize placeholders for combined results
        combined_abstract = []
        combined_cleaned_text = []

        # Split text into chunks and process each chunk
        for chunk in split_text(text, chunk_size=512):
            # Summarize the text chunk
            abstract = summarize_text(chunk)
            combined_abstract.append(abstract)

            # Clean the text chunk
            cleaned_text = clean_text(chunk)
            combined_cleaned_text.append(cleaned_text)

        # Combine results from all chunks
        final_abstract = ' '.join(combined_abstract)
        final_cleaned_text = ' '.join(combined_cleaned_text)

        # Append the data to the list
        data.append([title, final_abstract, final_cleaned_text])

    # Create a DataFrame from the data list
    df = pd.DataFrame(data, columns=['Title', 'Abstract', 'Content'])

    # Save the DataFrame to a CSV file
    output_file_path = 'processed_pdfs.csv'
    df.to_csv(output_file_path, index=False)

    return output_file_path
    
# Gradio interface
pdf_input = gr.File(label="Upload PDF Files", file_types=[".pdf"], file_count="multiple")
csv_output = gr.File(label="Download CSV")

gr.Interface(
    fn=process_pdfs, 
    inputs=pdf_input, 
    outputs=csv_output,
    title="Dataset creation",
    description="Upload PDF files and get a summarized CSV file.",
    article="""<p>This is an experimental app that allows you to create a dataset from research papers.</p>
                <p>This app uses the allenai/led-base-16384-multi_lexsum-source-long and sshleifer/distilbart-cnn-12-6 AI models.</p>
                <p>The output file is a CSV with 3 columns: title, abstract, and content.</p>"""
).launch(share=True)