On reconciling conflicting updates (a compromised revision approach)

This thesis proposes a method for handling logically con icting inputs into knowledge bases in a compromised way. The method revises nite bases which are subjected to some integrity constraints, and performs a reconciliation of the con icting data, by allowing the consistent logical consequences of the given up database elements or inputs, to update the database as a compromise. Since justi cations for those consequences are not available in the revised database, they are classi ed as non-supported consequences. Those non-supported data are viewed as extra information which survived from the revision process. Resolving con icting updates in dynamic databases, for instance, are frequent and critically important problems of real applications. Such problems require the revision of theories and knowledge bases. It is not realistic to aim for a generic approach in those cases, since theory revision is fundamentally dependent on application-speci c mechanisms, principles and heuristics. The approach we propose in this thesis, caters for the speci c case where compromise solutions for revising knowledge bases apply, when con icts involving updates occur. Our system considers that the information in the knowledge base is partially ordered. This ordering is supposed to contribute to the expressive power of the knowledge base representation, and to serve as basis for the construction of the safe-maximal consistent subsets which are introduced in this work, to provide a unique solution from the revision process. In comparison with approaches that require preference among con icting inputs, or that avoid them by cancelling them out completely, our approach ts as an alternative which provides more informative results. Examples of inputs include database updates, actions, and beliefs. In this thesis, we introduce the notion of reconciling con icting inputs with the knowledge base, by establishing some compromise reasoning policies. We characterize the compromise revision of our approach within the perspective of belief revision. In this characterization, we propose a base change operator r, which is based on a set of postulates de ned for compromise revision, considering nite bases with integrity constraints. We show that r is consistent and satis es a restricted version of the persistence notion. We also present a database updating system, denoted as CIULDS. This system deals with compromise updates in labelled databases, and is based on the Labelled Deductive Systems framework. In this formalization, the underlying framework's labelling facility is used to control the derivation process of the compromise consequences. The de ned labelling propagation conditions embed the control mechanism of our compromise-based approach. The update operations of CIULDS perform the reconciliation of the con icting inputs with the database, by invoking a compromise revision, denoted by the operator , whenever con icts arise. The revision operator is de ned on the basis of the revision operator r, however adapted to the updating requirements of CIULDS. The consequence relation of CIULDS is shown to satisfy particular versions of the properties of restricted re exivity, restricted monotonicity, the deduction theorem, strong transitivity (unitary cut), and non-explosiveness. Some of those properties follow also for the case that the conditional inclusion operation of the system is considered, in the place of the conventional set-theoretical inclusion. The updates and the transactions de ned for CIULDS, are shown to preserve the structural properties of the given initial labelled database, and to be consistent. As a complementing part of this work, we draw a simple comparison of our approach to con ict resolution, with some well known approaches in the area of reason maintenance systems.