Speaker
Description
Large Language Models (LLMs) have demonstrated considerable potential for ontology generation; however, their ability to capture the semantics of multi-level modelling remains insufficiently investigated. This work evaluates LLM-generated ontologies against the principles of Multi-Level Theory (MLT), using gold-standard ontologies from the literature as reference models. The evaluation focuses on four dimensions: abstraction-level correctness, meta-level representation, cross-level relationships, and syntactic validity. The results reveal recurring semantic errors, including confusion between classes and their instances, inconsistent assignment of entities to abstraction levels, and incorrect instantiation chains and cross-level relationships. Notably, these errors persist even in ontologies that are syntactically valid. Our findings expose limitations in the ability of current LLMs to understand and represent multi-level ontological structures, highlighting the need to improve their capacity to interpret abstraction levels, meta-level relationships, and the underlying semantics of multi-level modelling.
| Tlač postru/ Print poster | Budem požadovať tlač /I hereby required to print the poster in faculty |
|---|