Introduction
The Web Ontology Language (OWL) is fundamentally designed for the Open World Assumption (OWA). In this paradigm, missing information is treated as unknown rather than false. While beneficial for distributed web data, this makes OWL unsuitable for traditional data validation tasks. For instance, a constraint requiring every product to have a producer cannot be enforced if the reasoner simply infers that an unknown producer exists. Furthermore, the lack of a Unique Name Assumption (UNA) allows a reasoner to equate two distinct entities to satisfy a cardinality constraint.
Proposed Integrity Constraint Semantics
To bridge this gap, this research introduces a novel Integrity Constraint (IC) semantics for OWL. This approach treats axioms as epistemic constraints—questions about what the knowledge base knows rather than what is logically possible in some model.
A key component is the "Weak Unique Name Assumption," implemented through Minimal Equality (ME) models. In these models, distinct named individuals are assumed to be different unless explicitly stated otherwise. This prevents the "smushing" of entities during validation. The semantics focus specifically on named individuals, ignoring anonymous individuals inferred by the reasoner.
Implementation via SPARQL
The authors demonstrate that validating these ICs can be reduced to a query-answering problem. Specifically, OWL axioms are translated into Distinguished Conjunctive Queries with negation ($DCQ^{not}$). For common OWL fragments like $SRI$ and $SROI$, these queries can be executed directly using standard SPARQL with the NOT EXISTS pattern.
By treating constraints as "negative" queries (searching for violations), the system becomes reasoner-independent. If a query for a violation returns a result, the constraint is considered violated. This allows an ontology to contain both standard axioms for inference and IC axioms for validation.
Evaluation and Results
The prototype was evaluated using the LUBM(5) benchmark, comprising roughly 100,000 individuals and 800,000 ABox axioms. The results highlight a significant performance advantage: while a standard logical consistency check took an average of 10 seconds, validating a single IC via the SPARQL-based approach took only 2 seconds. This demonstrates that IC validation is not only more semantically appropriate for data quality tasks but also computationally efficient.
Conclusion and Future Work
This hybrid approach allows ontologies to serve a dual purpose: standard axioms provide open-world inferences, while IC axioms ensure data integrity through a closed-world, epistemic lens. Future work will focus on extending these proofs to the full $SROIQ$ logic (the basis of OWL 2) and optimizing the execution of multiple simultaneous constraints to improve scalability.
Summary updated