Explaining Ontology-Mediated Query Answers Using Proofs over Universal Models

Alrabbaa, Christian; Borgwardt, Stefan; Koopmann, Patrick; Kovtunova, Alisa

doi:10.1007/978-3-031-21541-4_11

Cited by 2 publications

(1 citation statement)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1 By "black-box" we mean that the details of the reasoning process are not relevant, and indeed any reasoner providing comparable reasoning services could be used. However, the semantics are completely transparent, and the results could be interpreted/explained using a wide variety of techniques from the extensive literature on this topic (e.g., [3]). Note that this differs from the definition of the term that is often used, e.g., in the context of Machine Learning (ML), where the semantics of the system is opaque to the user.…”

Section: Introductionmentioning

confidence: 99%

Conjunctive query answering over unrestricted OWL 2 ontologies

Igne

Germano

Horrocks

2023

View full text Add to dashboard Cite

Conjunctive Query (CQ) answering is a primary reasoning task over knowledge bases. However, when considering expressive description logics, query answering can be computationally very expensive; reasoners for CQ answering, although heavily optimized, often sacrifice expressive power of the input ontology or completeness of the computed answers in order to achieve tractability and scalability for the problem. In this work, we present a hybrid query answering architecture that combines various services to provide a CQ answering service for OWL. Specifically, it combines scalable CQ answering services for tractable languages with a CQ answering service for a more expressive language approaching the full OWL 2. If the query can be fully answered by one of the tractable services, then that service is used, to ensure maximum performance. Otherwise, the tractable services are used to compute lower and upper bound approximations. The union of the lower bounds and the intersection of the upper bounds are then compared. If the bounds do not coincide, then the “gap” answers are checked using the “full” service. These techniques led to the development of two new systems: (i) RSAComb, an efficient implementation of a new tractable answering service for RSA (role safety acyclic) (ii) ACQuA, a reference implementation of the proposed hybrid architecture combining RSAComb, PAGOdA, and HermiT to provide a CQ answering service for OWL. Our extensive evaluation shows how the additional computational cost introduced by reasoning over a more expressive language like RSA can still provide a significant improvement compared to relying on a fully-fledged reasoner. Additionally, we show how ACQuA can reliably match the performance of PAGOdA, a state-of-the-art CQ answering system that uses a similar approach, and can significantly improve performance when PAGOdA extensively relies on the underlying fully-fledged reasoner.

show abstract

Section: Introductionmentioning

confidence: 99%