A Theory of Complementarity for Extracting Accurate Data from Inaccurate Sources through Integration

The purpose of this research is to develop a theory that helps produce accurate data integration output given multiple, overlapping, inaccurate sources. At present, the theoretical foundations of solutions that center on source selection and conflict resolution with a similar goal are limited. This paper introduces a new solution approach and theory that center on notions of complementarity, based on the assumption that errors are not random. The essence of the new approach is the following: instead of concentrating on sources that are individually highly accurate, center on sources that have a complementary nature and yield highly accurate output when integrated. The theory can offer guides for the characterization of accuracy, effective use of accuracy estimates, source selection, and conflict resolution strategies. Implementations will require information about error patterns.