Ontology Debugging By Concept Subsumption

Description Logics are a family of knowledge representation languages, which areemployed in various areas, such as natural language processing, databases, biology, medicine,and geography, and the most notable success is that they are the theoretical basis of WebOntology Languages OWL today. Different OWL languages based on relative DLs languageswith different expressive power and complexity can fit different situations, with properfunctions and better performance.In the process of ontology construction, merging, maintenance, and query, it is difficult toavoid inference results out of the user’s expectations, and these unexpected inference resultsare more likely to produce with the scale and complexity increasing of ontologies. At thesame time, diagnosing the causes of reasoning results and repairing the ontology just rely onthe user experience is also becoming nearly impossible. The main method for solving suchproblems is ontology debugging, and ontology debugging generally includes two aspects,ontology diagnose and ontology repair. Ontology diagnosis generally gets ideas fromModel-based Diagnosis that, it usually finds out the minimal parts lead to inconsistent,provides information for user to repair ontology or automatically repairs the ontology bydirectly delete axioms in one minimal diagnosis. And ontology repair usually adds or removesaxiom according to the ontology based on the result of ontology diagnosis, to get in line withthe user expectations.Reasoning results do not conform to the user’s expectations generally fall into twocategories. On the one hand, the ontology may be inconsistent itself-there are some logicalcontradictions in it causing some unsatisfiable concept, and on the other hand, for an ontologywith no logical contradiction, the information of it may be not agreed with the domainknowledge. So far, the processes of ontology diagnosis for these two kinds of problems aregenerally converted into the standard reasoning problems concept satisfiablity. These methodare usually based on building conflict model by Tableau reasoning, eventually get a minimalaxiom collection (MUPS/Justification), but complex intermediate result of Tableau reasoningcan’t provide more useful information for users to find and deal with the second kinds ofunexpected problems.Compared with the concept satisfiability, concept subsumption is another kind of standardreasoning which computes the subsumption relation between concepts, and it can furtherproduce concepts hierarchy. The information is helpful for users to understand the concept ofontology structure easier, and valuable the second kinds of unexpected problems. Also,concept subsumption can be done by a good performance method "Consequence-based Reasoning", and do not need to be converted into concept satisfiability problem. Theintermediate results of "Consequence-based Reasoning" are in form of axioms and easy tounderstand and reuse. This paper discusses ontology debugging based on conceptsubsumption, with the following contents:(1) Describe a general process and give relative definition for ontology debugging,diagnosis and repair problems. Indicate that concept subsumption can provide much morevaluable additional information for ontology debugging than concept satisfiability based.Define inference tracing JProof closely related to Justifications based on Proof, and the JProofis more effective for dealing with the second kinds of unexpected problems.(2) Give a JProof construction method based on Consequence-based Reasoning rules forEL+, the method can also efficiently get all the Justification of corresponding problems at thesame time. Briefly introduce HS-Tree method for computing all minimal diagnoses.The experimental results show that this method not only can effectively solve JProof, andcan effectively compute all the Justifications. These features not only for EL+, also can adaptto the more complex DLs which can be handled by Consequences-based Reasoning, e.g. HornSROIQ, ALCI, even SHI.