Reducing Human Effort in Ontology-Based Data Integration: An LLM-Assisted Mapping Pipeline

Sulthoni Ashiddiiqi, Fajar J. Ekaputra, Muhammad Zuhri Catur Candra, Gusti Ayu Putri Saptawati Soekidjo

Keywords: Ontology-Based Data Integration, mapping, Large Language Models

Abstract: Ontology-Based Data Integration (OBDI) relies on semantic mappings between data sources and a target ontology to provide uniform access to heterogeneous data. However, creating these mappings is complex, time-consuming, and resource-intensive, often requiring significant human expertise. This paper introduces LAMP (LLM-Assisted Mapping Pipeline), a novel approach that leverages Large Language Models (LLMs) to assist the mapping process in OBDI. LAMP decomposes the mapping task into manageable subtasks, leveraging retrieval-augmented techniques, contextual sufficiency management, and coherence maintenance across subtasks to enhance LLM performance. The pipeline incorporates a human-in-the-loop component for quality assurance. We evaluate LAMP on the BLINKG dataset across three mapping scenarios, demonstrating significant improvements in F1-scores, particularly in complex mapping scenarios, compared to single-prompt approach. Our findings highlight LAMP’s potential to reduce human effort and improve efficiency in OBDI mapping tasks.

Submission Number: 9