ChatGPT_Law_PropertyGraph: Automated Analysis of Regulatory Documents (Diploma Thesis)

Overview

ChatGPT_Law_PropertyGraph is a modular Java-based pipeline that transforms legal PDF documents into structured knowledge graphs and compliance checklists. The system has been developed in the context of my Diploma Thesis and uses the EU AI Act as a primary case study.

The system produces:

• A property graph in GraphML format
• A Markdown checklist outlining obligations, rights, and prohibitions

---

High-level architecture of the legal document analysis pipeline.

Project Structure

The repository consists of three independent Maven projects, each located in its own subdirectory:

/LegislativeTextParser  
/LawToPropertyGraphGenerator  
/GraphMLToChecklist

Components at a Glance

• LegislativeTextParser: Extracts and cleans legal text from PDF files, preserving legal structure (Law → Chapter → Article → Paragraph).
• LawToPropertyGraphGenerator: Uses a Large Language Model (GPT-4o-mini) to extract legal entities and relationships, producing a property graph.
• GraphMLToChecklist: Generates a Markdown checklist summarizing legal responsibilities.

---

Pipeline Components

1. Legislative Text Parser (/LegislativeTextParser)

• Parses and cleans legal PDF documents.
• Extracts hierarchical structure: Law → Chapter → Article → Paragraph.
• Outputs:
  • law_structure.json
  • entities.txt to: src/resources/output/entities.txt

---

2. Law-to-Property-Graph Generator (/LawToPropertyGraphGenerator)

• Uses OpenAI GPT-4o-mini to extract legal entities and their verbal relations.
• Applies Levenshtein distance for entity normalization.
• Constructs a unified property graph.

Required Setup

• Create a file named env.properties in the root of this project with the following content:

  GPT_API_URL=https://api.openai.com/v1/chat/completions
  GPT_API_KEY=your_openai_key_here
  

• Copy entities.txt from the first project's output:

  From: /LegislativeTextParser/src/resources/output/entities.txt  
  To:   /LawToPropertyGraphGenerator/src/resources/output/entities.txt
  

• Output:

  • final.graphML in src/resources/output/

---

3. GraphML-to-Checklist (/GraphMLToChecklist)

• Loads the GraphML file into a JavaFX interface.
• Generates a Markdown checklist summarizing the graph nodes.

Usage Note

• On startup, load the following files:
  • GraphML:
    /LawToPropertyGraphGenerator/src/resources/output/final.graphML
  • JSON:
    Output from /LegislativeTextParser (e.g. law_structure.json)

---

Example Workflow

# 1. Run the legislative parser
cd LegislativeTextParser
mvn clean install
java -jar target/legislative-text-parser.jar

# 2. Copy the entities file to the second tool
cp src/resources/output/entities.txt ../LawToPropertyGraphGenerator/src/resources/output/

# 3. Add API credentials to env.properties in the second tool's root
# Create a file named "env.properties" in the project's root
# Paste inside :
GPT_API_URL=https://api.openai.com/v1/chat/completions
GPT_API_KEY=your_openai_key_here

# 4. Run the property graph generator
cd ../LawToPropertyGraphGenerator
mvn clean install
java -jar target/law-to-property-graph-generator.jar

# 5. Start the checklist generator and load the required files
cd ../GraphMLToChecklist
mvn clean install
java -jar target/graphml-to-checklist.jar

---

Technologies Used

• Java 21
• Maven
• JavaFX
• OpenAI GPT-4o-mini API
• GraphML (yEd-compatible)
• JSON & Markdown

---

Academic Context

This project was developed as part of the Diploma Thesis of
Papadopoulos Konstantinos, Undergraduate Student
Department of Computer Science and Engineering, University of Ioannina

Supervisor: Professor Panos Vasiliadis