SOEL: A New Exploration of Using Large Language Models to Assist Ontology Engineering

SOEL (Supporting Ontology Engineering with Large Language Models) is a project developed by the Ontology Engineering Group (OEG) at the Polytechnic University of Madrid. It aims to use large language models to provide intelligent support for ontology engineering, reducing the high cost and limited efficiency of manual modeling by domain experts in traditional ontology construction processes, and offering new intelligent tools for knowledge graph construction and semantic web technologies.

Ontology engineering is a core field of knowledge graph and semantic web technologies, involving the precise definition and organization of concepts, relationships, and rules. The traditional ontology construction process requires domain experts to invest a lot of time in manual modeling, which is costly and has limited efficiency. With the rapid development of large language models, researchers have begun to explore applying these models to various stages of ontology engineering to lower the threshold and improve efficiency. The SOEL project is a typical representative of this exploration direction.

The core goal of the SOEL project is to integrate the natural language understanding and generation capabilities of large language models with the structured knowledge representation of ontologies. The project provides services through a website, allowing users to complete multiple ontology engineering tasks with the help of large language models. The technical implementation may adopt the following strategies: First, use the semantic understanding ability of large language models to extract concepts and relationships from unstructured text; then assist users in ontology design, verification, and expansion through conversational interaction; finally, output the generated ontology in standard formats such as OWL and RDF for easy integration with other semantic web tools. This architecture not only retains the rigor of ontology engineering but also leverages the flexibility of LLMs, enabling non-professional users to participate in knowledge modeling.

The application value of SOEL is reflected in multiple aspects: For domain experts, it can accelerate the initial construction of ontologies and reduce repetitive work; for knowledge graph developers, it provides a rapid prototyping tool; for educators and researchers, it is an experimental platform for exploring human-machine collaborative knowledge engineering. Specific scenarios include: automatically extracting domain terms from document collections and suggesting concept hierarchies; automatically generating attribute definitions and constraints based on user descriptions; verifying the completeness and consistency of ontologies through natural language question answering; supporting the alignment and fusion of multilingual ontologies.

Although LLMs have broad application prospects in ontology engineering, they face several challenges: First, the accuracy issue—LLMs may generate ontology elements that seem reasonable but do not conform to domain knowledge, requiring manual review; second, consistency maintenance—it is difficult to ensure that new content is logically consistent with the existing structure as the ontology scale expands; third, the interpretability issue—users need to understand the basis of LLM recommendations to make decisions. SOEL needs to find a balance between automation and controllability.

Projects like SOEL may develop towards deeper integration in the future: On one hand, integrating with more ontology editing tools (such as Protégé) via plugins to improve practicality; on the other hand, combining Retrieval-Augmented Generation (RAG) technology to enable LLMs to perform more accurate reasoning based on existing ontology knowledge. In addition, introducing multimodal capabilities to identify conceptual relationships from tables, charts, and even images, enriching the data sources for ontology engineering is also a trend.

The SOEL project represents an important direction for the integration of artificial intelligence and traditional knowledge engineering, demonstrating how to use LLMs to lower the technical threshold of ontology engineering and enable more fields to benefit from semantic web technologies. As related technologies mature, we look forward to more similar tools emerging to promote the widespread application of knowledge graphs in various industries.