Towards LLM-augmented Creation of Semantic Models for Dataspaces

Dataspaces aim to enable smooth and reliable data exchange between different organizations. They have gained increasing attention in Europe following the enactment of the European Data Governance Act. This legislation emphasizes trust, accessibility, and shared dataspaces, which require semantic interoperability grounded in the FAIR principles. Although semantic descriptions in the form of semantic models and ontologies are integral to dataspaces, their full potential remains underutilized. Meaningful metadata, including contextual information, enhances data usability, but manually creating semantic models can be challenging. Large Language Models (LLMs) offer a new way to utilize data in dataspaces. Their advanced natural language processing capabilities enable context-aware data processing and semantic understanding. This paper presents initial experiments on customizing and optimizing LLMs for semantic labeling and modeling tasks. The contributions of this work include research questions for future investigations, early experiments demonstrating the applicability of LLM for semantic labeling, and proposed directions to address discovered challenges.