Abstract
In the Semantic Web, the Web Ontology Language (OWL) is widely used for its computational tractability, yet it lacks the ontological depth required for sophisticated conceptual modeling. This paper presents Scior, a tool that leverages 37 formal inference rules derived from the Unified Foundational Ontology (UFO) to automatically categorize "flat" OWL classes. By utilizing a small manual seed of classifications, Scior can exclude impossible categories or fully determine the ontological nature of remaining classes, bridging the gap between machine-processable logic and rich conceptual interoperability.
Introduction: The "Flatness" of OWL
While OWL is the gold standard for creating machine-processable data, it treats all classes as ontologically equivalent. Whether a class represents a fundamental type (e.g., Person), a temporary state (e.g., Student), or a shared property (e.g., Red), OWL provides no inherent mechanism to distinguish them. This lack of depth hinders "Purpose 1" of ontologies: creating a clear, shared conceptual model. Foundational ontologies like UFO provide these missing categories, but manually re-engineering millions of existing OWL classes is infeasible. Scior addresses this by automating the grounding of OWL classes into UFO categories.
Methodology
The researchers translated UFO axioms into 37 First-Order Logic inference rules focusing on two primary metaproperties:
- Rigidity: Determining if an entity belongs to a class for its entire existence or just temporarily.
- Sortality: Determining if a class provides a unique principle of identity for its members.
The Scior Workflow
- Seeding: A user manually classifies a small percentage (e.g., 10-20%) of classes in an OWL ontology.
- Inference: Scior applies the rules to infer categories for the remaining classes.
- Exclusion: The tool identifies impossible categories (e.g., a class cannot be a "Kind" if it specializes an anti-rigid type), significantly reducing the cognitive load for human modelers.
The rule set was formally verified for consistency using the Alloy modeling language.
Experimental Setup and Datasets
The tool was evaluated using the OntoUML/UFO Catalog. After filtering for size and consistency, the researchers produced two primary datasets:
- OWA Dataset: 85 taxonomies (1,624 classes) tested under Open World Assumptions.
- CWA Dataset: 67 taxonomies (1,234 classes) tested under Closed World Assumptions.
Key Results and Analysis
The effectiveness of Scior was measured by its ability to exclude potential categories based on varying levels of initial manual seeding (10% to 90%).
Open World vs. Closed World Assumptions
The tool's performance varies significantly based on the "world view" adopted. Under an Open World Assumption (OWA), the tool is conservative, assuming there may be unknown subclasses. Under a Closed World Assumption (CWA), which assumes the visible model is complete, the inference is much stronger.
Efficiency Metrics
- Low Seeding Impact: With only a 20% manual seed, Scior could exclude at least five out of eight possible categories for over 50% of the remaining classes in both OWA and CWA scenarios.
- Full Classification: Under OWA with a 90% seed, the tool fully and automatically classified approximately 20% of the remaining unknown classes.
- CWA Superiority: At a 50% seed level, CWA models resulted in at least some exclusions for 100% of the remaining classes, demonstrating the power of the tool when models are assumed to be complete.
Conclusion and Future Work
Scior demonstrates that foundational ontologies can be used to improve the quality of "web-friendly" OWL models through automated reasoning. While the system currently requires an initial manual seed and focuses primarily on independent taxonomies, future iterations could incorporate cross-taxonomy relationships (like associations) to further sharpen inference. Integrating Scior directly into ontology editors like Protégé remains a key goal for improving the practical workflow of ontology engineers.
Summary updated