Balancing Brave and Cautious Query Strategies in Ontology Debugging Patrick Rodler, Kostyantyn Shchekotykhin, Philipp Fleiss and Gerhard Friedrich ⋆ Alpen-Adria-Universität Klagenfurt Universitätsstrasse 65-67 9020 Klagenfurt, Austria firstname.lastname@aau.at Abstract. Sequential ontology debugging is aimed at the efficient discrimination between diagnoses, i.e. sets of axioms which must be altered or deleted from the ontology to restore consistency. By querying additional information the number of possible diagnoses can be gradually reduced. The selection of the best queries is crucial for minimizing diagnosis costs. If prior fault probabilities (FPs) are available, the best results are achieved by entropy based query selection. Given that FPs are only weakly justified, however, this strategy bravely suggests sub- optimal queries although more cautious strategies should be followed. In such a case, it is more efficient to follow a no-risk strategy which prefers queries that eliminate 50% of diagnoses independently of any FPs. However, choosing the appropriate strategy in advance is impossible because the quality of given pri- ors cannot be assessed before additional information is queried. We propose a method which combines advantages of both approaches. On the one hand, the method takes into account available meta information in terms of FPs and the user’s confidence in these. On the other hand, the method can cope with weakly justified FPs by limiting the risk of suboptimal query selections based on the user’s confidence in the FPs. The readiness to take risk is adapted depending on the outcome of previous queries. Our comprehensive evaluation shows that the proposed debugging method significantly reduces the number of queries com- pared to both the entropy based and the no-risk strategy for any choice of FPs. 1 Introduction Support of ontology development and maintenance is an important requirement for the extensive use of Semantic Web technologies. However, the correct formulation of log- ical descriptions is an error-prone task even for experienced knowledge engineers. On- tology debuggers [8, 3, 1] assist ontology development by identifying sets of axioms (called diagnoses) that have to be modified s.t. inconsistencies or unwanted entailments are avoided. In many cases the main problem of debugging is the big number of alter- native diagnoses. To solve the problem a set of heuristics to rank diagnoses was proposed by [4]. However, this solution is not satisfactory as it cannot be guaranteed that the top-ranked ⋆ The research project is funded by grants of the Austrian Science Fund (Project V-Know, con- tract 19996)