Smart Ontology Construction: Combining Human Expertise with LLMs for Smarter Business Taxonomies

Authors

  • Ramanathan Rajamani

Abstract

Ontologies are a fundamental piece in the development of semantic web technologies, artificial intelligence applications and knowledge representation systems. While the Large Language Models (LLMs) are very capable of automatically generating ontologies by detecting concepts, taxonomy structures and relationships, they still rely on human input because of the limitations of logical reasoning, hierarchical structuring, and contextual accuracy. To improve the quality, scalability, and domain relevance of LLM assisted ontology construction this research presents a Human in the Loop (HITL) framework. The study employs OSHA accident and injury data, data enhancement, exploratory data analysis (EDA), and structured prompt engineering with GPT-4 to automate ontology generation, expert review to fix logical inconsistencies, resolve ambiguities, and remove redundancy. Ontology quality, accuracy, completeness, relevance, and consistency are measured and the automated model starts with 0.69 accuracy, 2.46 relationships per record completeness, 0.78 relevance, which are all improved upon in human refinement. The improvements are validated quantitatively through paired t-tests showing statistically significant gains in accuracy, consistency, and relevance, and qualitatively through a structured expert questionnaire involving 15 domain experts across three evaluation rounds. This study also shows the practical benefit of domain specific taxonomies that have been shown to reduce operational decision latency, enhance machine learning prediction precision, and accelerate and affordably transform the digital. Taxonomy development is a core strategic capability that organizations with a mature taxonomy benefit from accelerated cloud migration, reduced integration costs, and significant annual savings. The research integrates human expertise and LLM-driven automation to provide a robust, data driven framework for enhancing decision making, improving operational efficiency, and creating a competitive advantage in the digital frontier as it lays the groundwork for further advances in reasoning capabilities, bias mitigation, and prompt engineering in the design of ontology.

Downloads

Published

2025-09-12

How to Cite

Rajamani, R. (2025). Smart Ontology Construction: Combining Human Expertise with LLMs for Smarter Business Taxonomies. Digital Repository of Theses - SSBM Geneva. Retrieved from https://repository.e-ssbm.com/index.php/rps/article/view/1003