Determining the relative accuracy of attributes

The relative accuracy problem is to determine, given tuples <i>t</i>₁ and <i>t</i>₂ that refer to the same entity <i>e</i>, whether <i>t</i>₁[<i>A</i>] is more accurate than <i>t</i>₂<i>A</i>, i.e., <i>t</i>₁<i>A</i> is closer to the true value of the <i>A</i> attribute of <i>e</i> than <i>t</i>₂<i>A</i>. This has been a longstanding issue for data quality, and is challenging when the true values of <i>e</i> are unknown. This paper proposes a model for determining relative accuracy. (1) We introduce a class of accuracy rules and an inference system with a chase procedure, to deduce relative accuracy. (2) We identify and study several fundamental problems for relative accuracy. Given a set <i>I</i>_e of tuples pertaining to the same entity <i>e</i> and a set of accuracy rules, these problems are to decide whether the chase process terminates, is Church-Rosser, and leads to a unique target tuple <i>t</i>_e composed of the most accurate values from <i>I</i>_e for all the attributes of <i>e</i>. (3) We propose a framework for inferring accurate values with user interaction. (4) We provide algorithms underlying the framework, to find the unique target tuple <i>t</i>_e whenever possible; when there is no enough information to decide a complete <i>t</i>_e, we compute top-<i>k</i> candidate targets based on a preference model. (5) Using real-life and synthetic data, we experimentally verify the effectiveness and efficiency of our method.