LLM Text Attribution Tool - Temporal Style Evolution Tracker

An unconventional LLM-based text attribution system using prompt engineering and RAG to assess text authenticity across diverse writing styles, with a focus on tracking style evolution over time.

Temporal Style Evolution Tracker (temporal_tracker.py)

Tracks how writing styles change over time:

• Version history analysis: Compare multiple versions of the same document
• Style drift detection: Identify significant changes in writing patterns
• Timeline visualization: Visual graphs showing authenticity scores and feature evolution

Features

• Store multiple versions of documents with timestamps
• Analyze authenticity scores and style features for each version
• Detect style drift between versions with configurable thresholds
• Generate timeline visualizations showing:
  • Authenticity score evolution
  • Style feature changes over time
  • Drift points marked on timeline
• Persistent storage of version history

Quick Start

python temporal_tracker.py

Usage Example

1. Add versions: Track different versions of a document

   > add
   Document ID: essay_1
   [Enter text...]
   

2. Analyze history: Get detailed analysis of style evolution

   > analyze
   Document ID: essay_1
   

3. Visualize timeline: Generate timeline graphs

   > visualize
   Document ID: essay_1
   

4. Detect drift: Find significant style changes

   > drift
   Document ID: essay_1
   

Requirements

See requirements.txt for dependencies.

Setup

1. Install dependencies:

   pip install -r requirements.txt

2. Set up API key (optional, uses fallback if not set):

   export OPENAI_API_KEY="your-key-here"

   Or create a .env file with:

   OPENAI_API_KEY=your-key-here
   

How It Works

1. Version Tracking: Each time you add a version, the system:

   • Analyzes text authenticity using optimized LLM prompts
   • Extracts style features (sentence length, lexical diversity, etc.)
   • Stores metadata with timestamp

2. Drift Detection: Compares consecutive versions to identify:

   • Significant changes in style features
   • Authenticity score shifts
   • Threshold-based drift points

3. Visualization: Creates multi-panel timeline graphs showing:

   • Authenticity score over time
   • Feature evolution
   • Drift detection points

Project Structure

├── temporal_tracker.py    # Main implementation
├── requirements.txt        # Dependencies
├── README.md              # This file
└── temporal_data/         # Storage directory (created automatically)
    └── history.json       # Version history database