Spaces:
Sleeping
Sleeping
| import streamlit as st | |
| import tensorflow as tf | |
| import numpy as np | |
| import nltk | |
| import os | |
| from nltk.tokenize import sent_tokenize | |
| from transformers import DistilBertTokenizerFast, TFDistilBertForSequenceClassification | |
| # 📁 Use safe cache directory inside Hugging Face or Docker | |
| os.environ["TRANSFORMERS_CACHE"] = "/tmp/huggingface" | |
| # 📥 Download NLTK tokenizer | |
| nltk_data_path = "/tmp/nltk_data" | |
| nltk.download("punkt_tab", download_dir=nltk_data_path) | |
| nltk.data.path.append(nltk_data_path) | |
| # 🔄 Load tokenizer and model from Hugging Face | |
| tokenizer = DistilBertTokenizerFast.from_pretrained( | |
| "distilbert-base-uncased", cache_dir="/tmp/huggingface" | |
| ) | |
| model = TFDistilBertForSequenceClassification.from_pretrained( | |
| "sundaram07/distilbert-sentence-classifier", cache_dir="/tmp/huggingface" | |
| ) | |
| # 🔮 Predict AI probability for a sentence | |
| def predict_sentence_ai_probability(sentence): | |
| inputs = tokenizer(sentence, return_tensors="tf", truncation=True, padding=True) | |
| outputs = model(inputs) | |
| logits = outputs.logits | |
| prob_ai = tf.sigmoid(logits)[0][0].numpy() | |
| return prob_ai | |
| # 📊 Analyze all sentences | |
| def predict_ai_generated_percentage(text, threshold=0.15): | |
| text = text.strip() | |
| sentences = sent_tokenize(text) | |
| ai_sentence_count = 0 | |
| results = [] | |
| for sentence in sentences: | |
| prob = predict_sentence_ai_probability(sentence) | |
| is_ai = prob <= threshold | |
| results.append((sentence, prob, is_ai)) | |
| if is_ai: | |
| ai_sentence_count += 1 | |
| total_sentences = len(sentences) | |
| ai_percentage = (ai_sentence_count / total_sentences) * 100 if total_sentences > 0 else 0.0 | |
| return ai_percentage, results | |
| # 🌐 Streamlit Web App | |
| st.set_page_config(page_title="AI Detector", layout="wide") | |
| st.title("🧠 AI Content Detector") | |
| st.markdown("This app detects the percentage of **AI-generated content** based on sentence-level analysis using DistilBERT.") | |
| # Initialize session state to avoid duplicates | |
| if "last_input" not in st.session_state: | |
| st.session_state.last_input = "" | |
| st.session_state.results = None | |
| st.session_state.percentage = None | |
| # 📋 User Input Area | |
| user_input = st.text_area("📋 Paste your text below to check for AI-generated sentences:", height=300) | |
| # 🔘 Analyze Button | |
| if st.button("🔍 Analyze"): | |
| if not user_input.strip(): | |
| st.warning("⚠️ Please enter some text to analyze.") | |
| else: | |
| # Store in session_state to avoid duplicates | |
| st.session_state.last_input = user_input | |
| ai_percentage, analysis_results = predict_ai_generated_percentage(user_input) | |
| st.session_state.results = analysis_results | |
| st.session_state.percentage = ai_percentage | |
| # Display only if results are present | |
| if st.session_state.results is not None: | |
| st.subheader("🔍 Sentence-level Analysis") | |
| for i, (sentence, prob, is_ai) in enumerate(st.session_state.results, start=1): | |
| label = "🟢 Human" if not is_ai else "🔴 AI" | |
| st.markdown(f"**{i}.** _{sentence}_\n\n → {label}") | |
| st.subheader("📊 Final Result") | |
| st.success(f"Estimated **AI-generated content**: **{st.session_state.percentage:.2f}%**") | |