Defining and Mining Functional Dependencies in Probabilistic Databases

Abstract: Functional dependencies -- traditional, approximate and conditional are of critical importance in relational databases, as they inform us about the relationships between attributes. They are useful in schema normalization, data rectification and source selection. Most of these were however developed in the context of deterministic data. Although uncertain databases have started receiving attention, these dependencies have not been defined for them, nor are fast algorithms available to evaluate their confidences. This paper defines the logical extensions of various forms of functional dependencies for probabilistic databases and explores the connections between them. We propose a pruning-based exact algorithm to evaluate the confidence of functional dependencies, a Monte-Carlo based algorithm to evaluate the confidence of approximate functional dependencies and algorithms for their conditional counterparts in probabilistic databases. Experiments are performed on both synthetic and real data evaluating the performance of these algorithms in assessing the confidence of dependencies and mining them from data. We believe that having these dependencies and algorithms available for probabilistic databases will drive adoption of probabilistic data storage in the industry.