Publication
SAMM Copilot: Bootstrapping Semantic Models with the Eclipse Semantic Modeling Framework from Domain Data in JSON Using Large Language Models
Nazanin Mashhaditafreshi; Andreas Textor; Pascal Rübel; Nastaran Moarefvand; Achim Wagner
In: 4th International Workshop on LLM-Integrated Knowledge Graph Generation from Text (Text2KG). International Workshop on LLM-Integrated Knowledge Graph Generation from Text (Text2KG-2025), June 1-5, Portoroz, Slovenia, CEUR Workshop Proceedings, 2025.
Abstract
The Semantic Aspect Meta Model (SAMM) is a modeling formalism for describing semantic models of parts of a Digital Twin – so-called Aspect Models. It is an open specification developed as part of the Eclipse Semantic Modeling Framework (ESMF). With SAMM being based on the Resource Description Framework (RDF) and Shapes Constraint Language (SHACL), Aspect Models are usually created and edited manually using a suitable textual editor or the graphical Aspect Model Editor. A well-defined mapping exists between Aspect Models and the JSON data they describe, enabling new bottom-up modeling approaches. In this way, instead of having a manual process for semantic modeling, Aspect Models can be automatically or semi-automatically derived from existing domain data in JSON format, making the modeling process more accessible and reducing manual effort. The proposed workflow translates JSON data into Aspect Models automatically using Large Language Models (LLMs). Our results demonstrate that LLMs can effectively bootstrap semantic models, and preliminary human evaluation suggests the feasibility and usefulness of this method in practice.