Learning Constraint Orderings for Direct Diagnosis

Given an inconsistent set of constraints, it is important to efficiently resolve the underly-ing conflicts. Conflict resolution can be regarded as a specific type of explanation, oftendenoted as diagnosis. Efficiency is of high importance in interactive constraint-based appli-cations. In this article, we utilize diagnosis knowledge from the past to improve diagnosisefficiency while also maintaining user-defined preference criteria. Our approach integratesmodel-based collaborative filtering with model-based diagnosis. This allows us to increasethe efficiency of diagnostic reasoning for determining preference-preserving diagnoses.With our experiments on real-world configuration knowledge bases (B2C, BUSYBOX,EA and LINUX KERNEL), we demonstrate an improved runtime performance comparedto state-of-the-art diagnosis approaches. At the same time, we are able to achieve a highaccuracy level in diagnosis prediction. With this, we also contribute to the growing bodyof literature on combining machine learning and constraint-based reasoning.