# OntoRAG: Enhancing Question-Answering through Automated Ontology Derivation from Unstructured Knowledge Bases

**Authors:** Yash Tiwari, Owais Ahmad Lone, Mayukha Pal  
**arXiv:** [2506.00664](https://arxiv.org/abs/2506.00664)  
**Date:** 2025-05-31  
**Summary page:** https://www.summarizepaper.com/en/paper/2506.00664v1/

## Key points

- Ontologies are essential for organizing knowledge bases and improving question answering systems driven by Large Language Models (LLMs)
- Traditional methods of creating ontologies rely heavily on manual input from domain experts, which is time-consuming and error-prone
- OntoRAG is an automated pipeline designed to extract ontologies from unstructured knowledge bases, focusing on electrical relay documents
- Utilizes techniques such as web scraping, PDF parsing, hybrid chunking, information extraction, knowledge graph construction, and ontology creation
- Incorporates LLMs and graph-based methods to enhance sensemaking capabilities
- Experimental results show that OntoRAG outperforms conventional approaches in terms of comprehensiveness and diversity
- Achieves an impressive comprehensiveness win rate of 85% against vector RAG and 75% against the best configuration of GraphRAG
- Represents a significant advancement towards realizing the vision of the semantic web by streamlining knowledge organization and retrieval processes while reducing reliance on manual efforts

## Summary

, , , , 

Ontologies are essential for organizing knowledge bases and improving the performance of question answering systems driven by Large Language Models (LLMs). However, traditional methods of creating ontologies rely heavily on manual input from domain experts, making it a time-consuming and error-prone process that is not practical for large and dynamic knowledge domains. To address this challenge, this paper introduces OntoRAG, an automated pipeline designed to extract ontologies from unstructured knowledge bases with a focus on electrical relay documents. Utilizing advanced techniques such as web scraping, PDF parsing, hybrid chunking, information extraction, knowledge graph construction, and ontology creation, OntoRAG converts unstructured data into a searchable ontology. By incorporating LLMs and graph-based methods, OntoRAG significantly enhances the overall sensemaking capabilities of the system. Experimental results demonstrate that OntoRAG surpasses conventional Retrieval Augmented Generation (RAG) and GraphRAG approaches in terms of comprehensiveness and diversity. Its effectiveness is evident in achieving an impressive comprehensiveness win rate of 85% against vector RAG and 75% against the best configuration of GraphRAG. This automated approach to ontology creation represents a significant advancement towards realizing the vision of the semantic web by streamlining knowledge organization and retrieval processes while reducing reliance on manual efforts by domain experts. Overall, this work contributes to advancing the field of question answering systems powered by LLMs and sets a new standard for automated ontology creation in large-scale knowledge domains.