Probabilistic Reasoning for Large Scale Databases

The complexity of probabilistic reasoning prohibits its application on a large scale of data. In order to reduce the complexity, implementations of modeling approaches restrict themselves with respect to expressive power or relax on the underlying probability theory. We present the implementation aspects of a probabilistic extension of stratified Datalog. This probabilistic deductive system is strictly based on the well-founded ground of probability theory. The prototypical implementation of the system handles the expensive computation of the probabilities separately from the reasoning process itself. Thus, we can use standard optimization strategies known from deterministic systems in order to cope with large amounts of data. By adding probabilistic reasoning to a deductive database system we gain the possibility of describing the information retrieval task as computing the probabilityP (d! q), i. e. the probability of the inference between a document d and a query q. Therefore, the logical view on databases plus a probabilistic generalization of the data model is a promising candidate for a breakthrough in integrating database and information retrieval technology on the way to multimedia information systems.