README.md

metadata

title: AIDetector
emoji: 📉
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: 5.45.0
app_file: app.py
pinned: false
license: mit

Advanced AI Text Detector 🔍

An advanced AI text detection system that identifies AI-generated content, particularly from ChatGPT and similar language models.

Features

🤖 Dual Detection Methods

• Transformer-based Detection: Uses fine-tuned RoBERTa model specifically trained on ChatGPT detection
• Statistical Analysis: Employs multiple linguistic metrics for robust detection

📊 Comprehensive Analysis Metrics

• Burstiness Analysis: Measures sentence length variation (human text is typically more "bursty")
• Vocabulary Diversity: Analyzes lexical richness and word variety
• Repetition Detection: Identifies repeated phrases and patterns
• Perplexity Scoring: Evaluates text predictability
• Punctuation Patterns: Analyzes punctuation consistency

🎯 High Accuracy Features

• Multi-method ensemble approach for improved accuracy
• Confidence scoring system
• Detailed explanations for each detection
• Visual probability distribution

How It Works

1. Input Processing: The text is tokenized and prepared for analysis
2. Transformer Analysis: If available, the RoBERTa model provides initial AI probability
3. Statistical Analysis: Multiple linguistic features are extracted and analyzed
4. Score Combination: Results are weighted and combined for final prediction
5. Result Generation: Detailed report with classification, confidence, and explanations

Detection Categories

• AI-Generated: >80% AI probability (High confidence)
• Likely AI-Generated: 60-80% AI probability (Medium confidence)
• Uncertain: 40-60% AI probability (Low confidence)
• Likely Human-Written: 20-40% AI probability (Medium confidence)
• Human-Written: <20% AI probability (High confidence)

Usage Tips

• Provide at least 100 words for optimal accuracy
• Longer texts generally yield more reliable results
• The detector works best with English text
• Results are probabilistic - use them as guidance, not absolute truth

Technical Stack

• Gradio: Interactive web interface
• Transformers: Hugging Face transformer models
• PyTorch: Deep learning backend
• SciPy/NumPy: Statistical analysis

Limitations

• Best performance with English text
• Requires sufficient text length (minimum 50 characters, optimal 100+ words)
• Detection accuracy may vary with highly technical or specialized content
• Should be used as a tool for guidance, not definitive judgment

Deployment

This app is designed to run on Hugging Face Spaces. Simply upload the files to your Space and it will automatically deploy.

Model Credit

This detector uses the Hello-SimpleAI/chatgpt-detector-roberta model from Hugging Face, combined with custom statistical analysis methods.

---

Note: AI detection is a rapidly evolving field. No detector is 100% accurate, and results should be interpreted with appropriate context and judgment.