Introduction
Ontology-Based Data Access (OBDA) provides a conceptual layer over existing data sources, allowing users to query information using high-level terms while the system leverages background knowledge to infer implicit facts. However, a significant gap exists between theoretical OBDA logics and the practical requirements of relational databases, which rely heavily on keys and functional dependencies. This paper introduces a framework using the $CFDI_{nc}^\forall$ logic to bridge this gap, offering a solution tailored for "SQL-literate" environments.
The $CFDI_{nc}^\forall$ Logic
The authors propose $CFDI_{nc}^\forall$ as a "sweet spot" logic. Unlike the common DL-Lite family, $CFDI_{nc}^\forall$ is designed to capture the nuances of relational schemas, including multi-attribute keys and path functional dependencies (PFDs).
This logic allows for the expression of complex constraints while maintaining computational feasibility. A key technical innovation is the handling of Inverse-Induced PFDs, ensuring that the interaction between inverse relationships and keys does not lead to an infinite expansion of inferred objects during the data completion phase.
A Combined Approach to Query Answering
To achieve efficient query answering, the framework employs a hybrid strategy that blends two traditional techniques:
- ABox Completion (Data Augmentation): The system enriches the physical data (ABox) with facts inferred from the ontology (TBox). This process operates with PTIME data complexity, ensuring scalability for large-scale databases.
- Query Rewriting: For information involving "anonymous" individuals—entities whose existence is implied by the ontology but not explicitly named in the database—the system rewrites user queries into complex SQL statements that account for these logical implications.
By leveraging primary keys already present in the database, the system often avoids the "object invention" (creation of Skolem constants) common in other OBDA approaches, making it highly compatible with standard SQL engines.
Experimental Evaluation
The system was evaluated using a modified Lehigh University Benchmark (LUBM), adapted with a relational schema to reflect real-world object-relational databases. Tests were conducted across scales from LUBM 10 to LUBM 1000.
The results demonstrate that the overhead for ABox completion is manageable, typically requiring about twice the time needed for initial data loading and indexing. On the largest dataset (LUBM 1000), most queries executed in under a second, with even the most complex rewritten queries completing within 73 seconds, proving the approach is viable for massive datasets.
Conclusion and Future Work
The $CFDI_{nc}^\forall$ framework offers a robust path for implementing OBDA in relational environments. By combining data completion with query rewriting, it achieves PTIME data complexity while supporting the expressive constraints necessary for modern data management. Future research directions include developing incremental data completion methods to handle dynamic updates and relaxing concept compatibility restrictions for more flexible schema mappings.
Summary updated