Estimating Range Queries using Aggregate Data with Integrity Constraints: a Probabilistic Approach

Abstract

The problem of recovering (count and sum) range queries over multidimensional data only on the basis of aggregate information on such data is addressed. This problem can be formalized as follows. Suppose that a transformation T producing a summary from a multidimensional data set is used. Now, given a data set D, a summary S=T(D) and a range query r on D, the problem consists of studying r by modelling it as a random variable defined over the sample space of all the data sets D' such that T(D) = S. The study of such a random variable, done by the definition of its probability distribution and the computation of its mean value and variance, represents a well-founded, theoretical probabilistic approach for estimating the query only on the basis of the available information (that is the summary S) without assumptions on original data.