Generate your own semantic web with LLMs

• Use it for Retrieval-Augmented Generation (RAG)
• Generate custom visualizations
• Xplainbale AI: explore semantic relations behind LLMs

Table of Contents

• About
• Features
• Public demo
• Usage
• Configuration
• Examples
• Contributing

About

The tool takes unstructured data (i.e., plain text) as input and generates a graph knowledge base using Large Language Models (LLMs). The resulting graph can be visualized or applied for Retrieval-Augmented Generation (RAG). The tool allows for generating graphs from unstructured data for the following use cases:

• Enhancing RAG systems by accessing graph databases instead of generating answers directly from unstructured data. Adding intermediate graph generation step improves overall RAG quality (see evidence here).
• Exploring the semantic web generated with LLMs. Knowledge graphs are easily visualized in 2-dimensional space, providing a tool for comprehensive overview of the resulting structures. They can be used to explore the strength of connections between inputs and outputs in LLMs contributing to the field of Explainable AI (XAI) (see overview of applying graphs to XAI).
• Generating graphs for data exploratory analysis. The tool allows for generating semantic web from unstructured data for research purposes, such as data science, linguistics analysis or open domain data visualization. For example, such a tool can visualize topics and connections between them in a graph, providing a more agile alternative to bibliometric exploratory tools such as VOSViewer, or linguistic research ontologies, like WordNet.

Features

• Convert Unstructured Text to Knowledge Graph
  • Transform plain text (e.g., articles, reports, notes) into structured graph representations.
  • Extract entities, relationships, and semantic hierarchies.
• LLM-Powered Graph Construction
  • Test and apply various Large Language Models (e.g., DeepSeek, LLaMA) for semantic parsing and relationship extraction.
  • Generates knowledge graphs to enhance and explain RAG pipelines by providing and visualizing structured context.
• Interactive Visualization
  • Intuitive interface for exploring and manipulating generated graphs.
  • Node-link diagrams with zoom, filter, and search capabilities.

Public Web Demo

Accessible at https://xsynthia.onrender.com/ for testing and validation.

Usage

Start the application from a terminal:

git clone git@github.com:vifirsanova/xSynthia.git
cd xSynthia
./run.sh # Run script and start the application on your browser

Prerequisites:
Python 3.8+

Configuration

To run the application locally, provide your configuration for Groq or HuggingFace API:

# config.toml
api_key = "your_api_key_here"  # Replace with your Groq or HuggingFace API key