Approximate Denial Constraints

TL;DR

This paper introduces ADCMiner, an algorithm for mining approximate denial constraints from data, allowing for more flexible and accurate integrity constraint discovery in inconsistent databases by considering various definitions of approximation.

Contribution

The paper presents ADCMiner, a flexible algorithm that can incorporate different semantics of approximate denial constraints without relying on a single predefined approximation definition.

Findings

ADCMiner effectively mines approximate DCs with high accuracy.

The algorithm reduces runtime through sampling techniques.

It supports multiple definitions of approximation, enhancing versatility.

Abstract

The problem of mining integrity constraints from data has been extensively studied over the past two decades for commonly used types of constraints including the classic Functional Dependencies (FDs) and the more general Denial Constraints (DCs). In this paper, we investigate the problem of mining approximate DCs (i.e., DCs that are "almost" satisfied) from data. Considering approximate constraints allows us to discover more accurate constraints in inconsistent databases, detect rules that are generally correct but may have a few exceptions, as well as avoid overfitting and obtain more general and less contrived constraints. We introduce the algorithm ADCMiner for mining approximate DCs. An important feature of this algorithm is that it does not assume any specific definition of an approximate DC, but takes the semantics as input. Since there is more than one way to define an approximate DC and different definitions may produce very different results, we do not focus on one definition, but rather on a general family of approximation functions that satisfies some natural axioms defined in this paper and captures commonly used definitions of approximate constraints. We also show how our algorithm can be combined with sampling to return results with high accuracy while significantly reducing the running time.