Leveraging discarded samples for tighter estimation of multiple-set aggregates

Many datasets, including market basket data, text or hypertext documents, and events recorded in different locations or time periods, can be modeled as a collection of sets over a ground set of keys. Common queries over such data, including similarity or association rules are represented as the weight or selectivity of keys that satisfy some selection predicate defined over keys' attributes and memberships in particular sets. On massive data sets, exact computation of such aggregates can be inefficient or infeasible, and therefore, approximate queries are processed over sketches of the sets. Sketches based on coordinated random samples are scalable and flexible and well suited for many applications. Queries are resolved by producing a sketch of the union of the sets used in the predicate, from the sketches of these sets, and then applying an estimator to this union-sketch. We derive novel tighter (unbiased) estimators that leverage sampled keys that are present in the union of applicable sketches but excluded from the union sketch. We establish analytically that our estimators dominate estimators applied to the union-sketch for all queries and data sets. Empirical evaluation on synthetic and real data reveals that on typical applications we can expect a 25%-75% reduction in estimation error.