User Feedback in Probabilistic XML

Data integration is a challenging problem in many application areas. Approaches mostly attempt to resolve semantic uncertainty and conflicts between information sources as part of the data integration process. In some application areas, this is impractical or even prohibitive, for example, in an ambient environment where devices on an ad hoc basis have to exchange information autonomously. We have proposed a probabilistic XML approach that allows data integration without user involvement by storing semantic uncertainty and conflicts in the integrated XML data. As a consequence, the integrated information source represents all possible appearances of objects in the real world, the so-called possible worlds. In this paper, we show how user feedback on query results can resolve semantic uncertainty and conflicts in the integrated data. Hence, user involvement is effectively postponed to query time, when a user is already interacting actively with the system. The technique relates positive and negative statements on query answers to the possible worlds of the information source thereby either reinforcing, penalizing, or eliminating possible worlds. We show that after repeated user feedback, an integrated information source better resembles the real world and may converge towards a non-probabilistic information source.