Optimization Methods for Logic-based Query Answering from Inconsistent Data Integration Systems

Information integration systems providing the user with transparent access to heterogeneous data sources through a unified global view of all data, have emerged as a crucial issue in many application domains. This global view usually comprises integrity constraints which should be satisfied by the data retrieved from the sources. However, they might be often violated, and suitable approaches for handling such a situation are needed, taken inconsistency and incompleteness of source data into account. To this end, several recent works reduce query answering in data integration systems to non-monotonic logic programming. This branch of computational logic is well-suited for dealing with inconsistent and incomplete information, and offers the expressiveness which is needed to model query answering, which is in the worst case intractable. Furthermore, advanced implementations of systems including stable model engines like DLV and Smodels for evaluation of non-monotonic logic programs are available. Nonetheless, a naive reduction of data integration to non-monotonic logic programs is infeasible for larger data sets, and calls for optimizations. We address this issue and present several techniques of different nature which make a non-monotonic logic programming approach effective. The first is a technique to prune a logic program for query answering such that only the relevant rules are kept. The second is a sophisticated technique, developed at a generic level, which aims at localizing inconsistency and “repairing” it in a decomposition manner. Finally, we present a technique to combine non-monotonic logic programming and commercial relational database engines in order to facilitate efficient query answering from repairs computed for an inconsistent global data view. Our methods are not bound to a specific non-monotonic logic programming system. In particular, the results on localization and decomposition are independent of a logic program approach, and may be exploited for efficient realization of advanced data integration systems in general.