Privacy Policy Clause Classifier

This project provides a web interface for classifying sections of Privacy Policies and Terms of Service documents using a fine-tuned BERT model. It consists of a FastAPI backend for model inference and a Streamlit frontend for user interaction.

Components

Frontend (frontend_streamlit.py)

A Streamlit-based web interface that allows users to:

• Input sections of privacy policies or terms of service
• Submit text for classification
• View classification results with confidence scores
• Visualize predictions through interactive bar charts

Backend (backend_fastapi.py)

A FastAPI server that:

• Loads a custom fine-tuned BERT model for multi-label classification
• Processes incoming text requests
• Returns predicted categories with confidence scores
• Handles model inference using PyTorch

Setup

1. Install the required packages:

pip install -r requirements.txt

2. Start the backend server:

uvicorn backend_fastapi:app --reload

3. Launch the frontend:

streamlit run frontend_streamlit.py

Usage

1. Open the Streamlit interface in your browser (typically http://localhost:8501)
2. Paste a section of privacy policy or terms of service into the text area
3. Click "Classify Section" to analyze the text
4. View the results, including:
   • Detected privacy categories
   • Confidence scores
   • Visual representation of all category probabilities

Model

The system uses the Hacktrix-121/bert-base-uncased-opp115-multilabel model, which is trained on the OPP-115 Privacy Policy Dataset to identify 12 different privacy policy categories:

• First Party Collection/Use
• Third Party Sharing/Collection
• User Choice/Control
• User Access, Edit & Deletion
• Data Retention
• Data Security
• Policy Change
• Do Not Track
• International & Specific Audiences
• Miscellaneous and Other
• Contact Information
• User Choices/Consent Mechanisms

The model is fine-tuned on BERT base uncased for multi-label classification of privacy policy segments. The OPP-115 dataset consists of 115 privacy policies (267K words) annotated by skilled annotators for various privacy practice categories.

Technical Requirements

• Python 3.7+
• FastAPI for the backend server
• Streamlit for the web interface
• PyTorch and Transformers for model inference
• See requirements.txt for complete dependencies