Leveraging Large Language Models for Ontology-Based Data Access: A Preliminary Analysis

Abstract

In Ontology-Based Data Access (OBDA), we study how to represent legacy data sources using ontologies. This enables a modern, distributed, uniform data representation format with the ability to perform intelligent querying and processing. This task requires the development of software to interpret the data and express it as ontologies, which takes considerable time. On the other hand, large language models (LLM) have lately shown themselves to be great solution providers due to their ability to generate solutions from input specified in natural language by an end user. In this paper, we explore the potential of LLM to perform OBDA automatically. Our research hypothesis is that is possible to use an LLM tool like ChatGPT to perform OBDA. For this purpose, we studied ChatGPT responses with different problems associated with OBDA. We discovered that ChatGPT is able to generate ontologies from free text as well as from tables expressed as text or in CSV format. ChatGPT is also able to generate SPARQL queries, and it is also successful in expressing relational tables as ontologies being capable of correcting violations of integrity constraints when appropriately directed.