Data Exchange in Datalog Is Mainly a Matter of Choice

Abstract

Data exchange is the problem of migrating a data instance from a source schema to a target schema such that the materialized data on the target schema satisfies the integrity constraints specified by: TGDs (Tuple Generating Dependencies), which are universal quantified formulas with additional existential quantifiers, and EGDs (Equality Generating Dependencies), which are universal quantified formulas enforcing the equality of two variables. This paper presents a formulation of the data exchange problem using DATALOG with choice, which is a non deterministic construct based on stable model semantics. TGDs are represented by rules and a choice predicate is used to non-deterministically select values for the existential variables. Every EGD can be naturally represented by a goal rule. However, as in general it expresses a functional dependency, in this case the goal rule can be replaced by a choice predicate defining the functional dependency inside one of TGD rules. Although classical certainty semantics for query answering in a data exchange setting can be also defined for DATALOG with choice, this paper explores another direction: searching for a solution for which a number of given “sensible” queries have uncertainty-guaranteed answers. The paper discusses properties of privacy-preserving data exchange and illustrates its complexity. Finally, EGDs are extended to express count constraints (e.g, an employee may manage at most k departments instead of only one) and the choice construct is therefore extended to implement count constraints. The resulting setting can be used to define the exchange of aggregate data.