Centrala begrepp
Large Language Models (LLMs) can be effectively leveraged to automate the challenging task of complex ontology alignment, which involves detecting sophisticated 1-to-n, n-to-1, or m-to-n equivalence or subsumption relationships between ontology concepts.
Sammanfattning
This paper investigates the application of Large Language Models (LLMs) to tackle the complex ontology alignment challenge. The authors leverage a prompt-based approach and integrate rich ontology content, known as "modules", to significantly advance the automation of complex alignment tasks.
The paper first provides an overview of ontology alignment, highlighting the limitations of traditional approaches in detecting complex alignments. It then discusses the potential of LLMs, driven by advancements in Natural Language Processing (NLP), to enhance ontology engineering practices, including ontology alignment.
The core of the paper focuses on the design of the prompting process, which involves leveraging various techniques such as zero-shot, few-shot, and chain-of-thought prompting to guide the LLM in identifying relevant ontology components. The authors emphasize the importance of incorporating ontology module information, which provides additional context and structure, to improve the LLM's performance in detecting complex alignments.
The evaluation section presents a detailed analysis of the authors' approach using the GeoLink complex alignment dataset. The results demonstrate that the inclusion of module information significantly enhances the LLM's ability to identify the relevant ontology components, with high recall and precision rates in many cases. The authors also discuss the challenges encountered, such as the difficulty in detecting alignments involving type or class relationships, and provide insights into the impact of module richness on the discovery process.
The paper concludes by highlighting the significance of their contribution, which presents the first reasonably working approach for generating complex alignments without relying on shared individuals, a realistic setting often lacking in practical applications. The authors also outline future research directions, including the exploration of alternative module representations, fine-tuning of LLMs, and the integration of symbolic data and algorithms to further improve the complex ontology alignment task.
Statistik
The GeoLink dataset contains 109 complex alignment rules between the GeoLink Base Ontology (GBO) and the GeoLink Modular Ontology (GMO).
Citat
"Ontology alignment has been studied for over two decades, resulting in the development of many alignment approaches and systems. The majority of these systems are designed to detect only so-called "simple" 1-to-1 mappings between ontologies, primarily by establishing equivalence relationships between classes (unary predicates), or between properties (binary relationships)."
"With significant advancements in the natural language processing (NLP) and natural language understanding (NLU) fields, spurred by Large Language Models (LLMs), it has become possible to extract meanings from text and reason about it more effectively."