Supporting Probabilistic Data in a Relational System

Abstract:

It is often desirable to represent in a database entities whose properties cannot be deterministically classified. We develop a new data model that includes probabilities or confidences associated with the values of the attributes. Thus we can think of the attributes as random variables with probability distributions dependent on the entity the tuple purportedly describes. This new model offers a richer descriptive language allowing the database to more accurately reflect the uncertain real world. It also
offers a new interpretation of information incompleteness. We study three sets of issues: the proper model for probabilistic data, the semantics of probabilistic data, and the choice of operators and language necessary to manipulate such data.