app.py

import os
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
import torch
import spacy
import subprocess
import nltk
from nltk.corpus import wordnet
from gensim import downloader as api

# Ensure necessary NLTK data is downloaded
nltk.download('wordnet')
nltk.download('omw-1.4')

# Ensure the SpaCy model is installed
try:
    nlp = spacy.load("en_core_web_sm")
except OSError:
    subprocess.run(["python", "-m", "spacy", "download", "en_core_web_sm"])
    nlp = spacy.load("en_core_web_sm")

# Load a smaller Word2Vec model from Gensim's pre-trained models
word_vectors = api.load("glove-wiki-gigaword-50")

# Check for GPU and set the device accordingly
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load AI Detector model and tokenizer from Hugging Face (roberta-base-openai-detector)
tokenizer_ai = AutoTokenizer.from_pretrained("roberta-base-openai-detector")
model_ai = AutoModelForSequenceClassification.from_pretrained("roberta-base-openai-detector").to(device)

# AI detection function using the RoBERTa-based model
def detect_ai_generated(text):
    inputs = tokenizer_ai(text, return_tensors="pt", truncation=True, max_length=512).to(device)
    with torch.no_grad():
        outputs = model_ai(**inputs)
        probabilities = torch.softmax(outputs.logits, dim=1)
    ai_probability = probabilities[0][1].item() * 100  # Probability of being AI-generated
    human_probability = 100 - ai_probability  # Probability of being Human-written
    
    # Determine the label based on the higher probability
    if ai_probability > human_probability:
        label = "AI"
        probability = ai_probability
    else:
        label = "Human"
        probability = human_probability
    
    return f"The content is {probability:.2f}% {label} Written", probability

# Function to get synonyms using NLTK WordNet
def get_synonyms_nltk(word, pos):
    synsets = wordnet.synsets(word, pos=pos)
    if synsets:
        lemmas = synsets[0].lemmas()
        return [lemma.name() for lemma in lemmas]
    return []

# Function to capitalize the first letter of sentences and proper nouns
def capitalize_sentences_and_nouns(text):
    doc = nlp(text)
    corrected_text = []

    for sent in doc.sents:
        sentence = []
        for token in sent:
            if token.i == sent.start:  # First word of the sentence
                sentence.append(token.text.capitalize())
            elif token.pos_ == "PROPN":  # Proper noun
                sentence.append(token.text.capitalize())
            else:
                sentence.append(token.text)
        corrected_text.append(' '.join(sentence))

    return ' '.join(corrected_text)

# Paraphrasing function using SpaCy and NLTK
def paraphrase_with_spacy_nltk(text):
    doc = nlp(text)
    paraphrased_words = []
    
    for token in doc:
        # Map SpaCy POS tags to WordNet POS tags
        pos = None
        if token.pos_ in {"NOUN"}:
            pos = wordnet.NOUN
        elif token.pos_ in {"VERB"}:
            pos = wordnet.VERB
        elif token.pos_ in {"ADJ"}:
            pos = wordnet.ADJ
        elif token.pos_ in {"ADV"}:
            pos = wordnet.ADV
        
        synonyms = get_synonyms_nltk(token.text.lower(), pos) if pos else []
        
        # Replace with a synonym only if it makes sense
        if synonyms and token.pos_ in {"NOUN", "VERB", "ADJ", "ADV"} and synonyms[0] != token.text.lower():
            paraphrased_words.append(synonyms[0])
        else:
            paraphrased_words.append(token.text)
    
    # Join the words back into a sentence
    paraphrased_sentence = ' '.join(paraphrased_words)
    
    # Capitalize sentences and proper nouns
    corrected_text = capitalize_sentences_and_nouns(paraphrased_sentence)
    
    return corrected_text

# Combined function: Paraphrase -> Capitalization
def paraphrase_and_correct(text):
    # Step 1: Paraphrase the text
    paraphrased_text = paraphrase_with_spacy_nltk(text)
    
    # Step 2: Capitalize sentences and proper nouns
    final_text = capitalize_sentences_and_nouns(paraphrased_text)
    
    return final_text

# Gradio interface definition
with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            t1 = gr.Textbox(
                lines=5, 
                label='Text',
                value="There are a few things that can help protect your credit card information from being misused when you give it to a restaurant or any other business:\n\nEncryption: Many businesses use encryption to protect your credit card information when it is being transmitted or stored. This means that the information is transformed into a code that is difficult for anyone to read without the right key."
            )
            button1 = gr.Button("🤖 Predict!")
            label1 = gr.Textbox(lines=1, label='Predicted Label 🎃')
            score1 = gr.Textbox(lines=1, label='Probability (%)')

    button1.click(detect_ai_generated, inputs=[t1], outputs=[label1, score1])

demo.launch()