Q-ACER: Query Aggregate Constraint Efficient Repair System

Abstract

Selection processes, e.g., determining qualified job candidates or picking vendors, can naturally be modeled as relational queries. To ensure that such a query produces legally compliant and ethically sound results, the outputs of the query are often subject to additional constraints including fairness ratios, budget limits, and coverage requirements. When such constraints are violated, users are typically left without guidance on how to repair their queries. In this demonstration, we present Q-ACER (Query Aggregate Constraint Efficient Repair System), an efficient query repair system that enumerates candidate repairs of an input query such that the result set of the updated query fulfills user-provided constraints. In comparison to other query repair frameworks, Q-ACER covers a significantly larger class of constraints which is necessary for supporting fairness metrics such as statistical parity. However, this necessitates the development of novel techniques for sharing computations when evaluating constraints as well as evaluating multiple query repair candidates at once.