Norouzi et al. 2025 – Semantic Representation of Processes with Ontology Design Patterns
Citation
Norouzi, E., Hertling, S., Waitelonis, J., & Sack, H. (2025). Semantic Representation of Processes with Ontology Design Patterns. SeMatS 2025: 2nd International Workshop on Semantic Materials Science, ISWC 2025, Nara, Japan.
Core Argument
Ontology Design Patterns (ODPs) offer modular, reusable solutions for recurring modeling problems, but they’re often implicit, embedded in ontologies without being explicitly documented or published. This limits reuse by domain experts who would benefit from these patterns but can’t navigate entire ontologies to find them.
Key Contributions
- Survey of ontologies for scientific workflow and process modeling (PMDcore, GPO, WILD, P-PLAN, Metadata4Ing, OPMW)
- Three-requirement framework for evaluating process representation: Process Structure, Data/Resources, Project Context
- Baseline method for automatic ODP extraction using semantic similarity
- Evaluation against curated ground truth patterns
Method
Uses sentence transformer embeddings to match natural language requirements against ontology annotations (rdfs:label, skos:definition, rdfs:comment). Computes cosine similarity, applies threshold, evaluates with precision/recall/F1.
Key Finding
P-PLAN achieved highest F1 scores for Process and Resource ODPs. PMDcore was the only ontology supporting all three requirement categories. The approach shows promise but has bias toward ontologies with richer textual annotations.
Limitations Acknowledged
- Annotation quality bias (better labels = higher scores)
- Requirement formulation affects results
- Processes each ontology independently (can’t find cross-ontology patterns)
- Modularization is only one path to reuse
Resources
GitHub: https://github.com/ISE-FIZKarlsruhe/odps4mse