Text2KG

This repository contains the implementation for Text2KG, a research project focused on automated knowledge graph (KG) generation from unstructured data using three prominent language models: GPT-4, LLaMA 2 (13B), and BERT. The goal is to simplify KG creation for GraphRAGs (Graph-based Retrieval-Augmented Generative Systems) and evaluate the models based on several performance metrics.

Features

• Automated Knowledge Graph Creation: Generate KGs directly from raw text without manual relationship classification.
• Multi-Model Support: Scripts for GPT-4, LLaMA 2, and BERT are provided.
• Performance Evaluation: Includes metrics such as Precision, Recall, F1-Score, Graph Edit Distance, and Semantic Similarity for comparing the generated KGs.
• Dataset: Contains the Wikipedia excerpt on C programming language used as input data.
• Visualization: Generates visual representations of KGs for all three models.

Repository Structure

.
├── data/                  # Input dataset (Wikipedia excerpt)
├── models/                # Scripts for GPT-4, LLaMA 2, and BERT
│   ├── gpt4.py            # GPT-4 KG generation script
│   ├── llama2.py          # LLaMA 2 KG generation script
│   └── bert.py            # BERT KG generation script
├── metrics/               # Scripts to calculate evaluation metrics
├── visualizations/        # Generated KG visualizations
├── results/               # Output tables and evaluation results
├── README.md              # Project overview (this file)
└── requirements.txt       # Dependencies for the project

Usage

1. Clone the repository:

   git clone https://github.com/AhanBhatt/Text2KG.git
   cd Text2KG

2. Install dependencies:

   pip install -r requirements.txt

3. Run KG generation:

   • To generate KGs using GPT-4:

     python models/gpt4.py

   • To generate KGs using LLaMA 2:

     python models/llama2.py

   • To generate KGs using BERT:

     python models/bert.py

4. Visualize and Evaluate:

   • Visualizations will be saved in /visualizations/.
   • Evaluation metrics will be stored in /results/.

---

Key Highlights

• Provides a systematic comparison of GPT-4, LLaMA 2, and BERT for KG creation.
• Reproducible methodology with pre-written scripts for all three models.
• Evaluation focuses on precision, accuracy, and structural fidelity of KGs.

---

Applications

This project is a step toward automating KG creation for GraphRAGs and has applications in:

• Information retrieval
• Knowledge representation
• AI-driven reasoning systems

---

Contributing

Feel free to contribute by extending the dataset, optimizing model implementations, or suggesting additional evaluation metrics.

---