Monday 07 April 2025
The quest for a more efficient way to create ontologies, complex systems of organized knowledge, has long been a challenge in the field of artificial intelligence. Ontologies are essential for tasks like question-answering and natural language processing, but building them can be a laborious and time-consuming process.
Recently, researchers have turned to large language models (LLMs) as a potential solution. These models are trained on vast amounts of text data and can generate human-like language with ease. But can they also create ontologies?
To find out, a team of scientists experimented with using LLMs to generate ontologies for three different domains: music, hospital care, and theatre. They created user stories and competency questions (CQs) to serve as guidelines for the models, and then fed these into the LLMs.
The results were promising. The LLMs were able to create ontology drafts that accurately reflected the CQs and user stories, with some even producing complete and correct ontologies. But there was a catch: many of the generated ontologies contained superfluous elements, like unnecessary classes and properties. This could make them difficult to use in practical applications.
To address this issue, the researchers developed two new prompting techniques for the LLMs. The first, called Memoryless CQbyCQ, reduces the input context size of the model to enhance performance and reduce cost. The second, Ontogenia, uses a carefully crafted prompt to encourage the model to produce more accurate and relevant output.
The results were impressive. When tested against three commercial LLMs, the new prompting techniques significantly improved the quality of the generated ontologies. One model, OpenAI o1- preview, produced ontology drafts that were comparable to those created by human experts in terms of accuracy and completeness.
But what does this mean for the future of artificial intelligence? If we can develop LLMs that are capable of generating high-quality ontologies with minimal human intervention, it could revolutionize the way we approach tasks like question-answering and natural language processing. It could also open up new possibilities for applications like expert systems, decision support systems, and more.
Of course, there are still many challenges to overcome before LLMs can be widely used for ontology generation. But the potential benefits are significant, and researchers are eager to continue exploring this exciting new frontier in AI.
Cite this article: “Unlocking the Potential of Large Language Models: A Study on Ontology Generation and Evaluation”, The Science Archive, 2025.
Large Language Models, Ontologies, Artificial Intelligence, Question-Answering, Natural Language Processing, Music, Hospital Care, Theatre, User Stories, Competency Questions
