app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModel, AutoModelForSequenceClassification
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 GPT-3.5-turbo model and tokenizer from Hugging Face
tokenizer_ai = AutoTokenizer.from_pretrained("Xenova/gpt-3.5-turbo")
model_ai = AutoModel.from_pretrained("Xenova/gpt-3.5-turbo").to(device)

# AI detection function using GPT-3.5-turbo-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)
        # Since this model does not directly output classification logits, you'll need to process the hidden states
        # For simplicity, let's just use the first hidden state for now (you may need to adjust based on your use case)
        hidden_state = outputs.last_hidden_state[:, 0, :]  # Use the first token's representation
        # Example: calculate some kind of score based on the hidden state
        score = torch.mean(hidden_state).item()
    return f"AI-Generated Content Score: {score:.2f}"

# 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 interface:
    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(lines=5, label="Input Text")
            detect_button = gr.Button("AI Detection")
            paraphrase_button = gr.Button("Paraphrase & Correct")
        with gr.Column():
            output_text = gr.Textbox(label="Output")

    detect_button.click(detect_ai_generated, inputs=text_input, outputs=output_text)
    paraphrase_button.click(paraphrase_and_correct, inputs=text_input, outputs=output_text)

# Launch the Gradio app
interface.launch(debug=False)