Repairing integrity rules for improved data quality

Abstract

Integrity constraints are the primary tool used to capture business rules and domain constraints in data management systems. When these constraints are not strictly enforced, poor data quality often arises, as inconsistencies occur between the data and the set of constraints. To resolve these inconsistencies, organisations often implement specific, sometimes manual, cleansing routines to fix the errors. As modern systems are expected to handle increasing amounts of highly heterogeneous data, often in dynamic data environments where the data and the constraints may change, manual cleansing routines are insufficient to handle this increased scale and heterogeneity. In this work, we present a set of new constraint repair operations that can be incorporated into a data quality tool that provides automated support for both data and constraint repair and management. Our holistic approach is designed to facilitate the curation and maintenance of both the data and the constraints. We focus on discovering trends, contextual information, and data patterns to understand how a business rule (constraint) has evolved. We also investigate how to find a minimal set of constraints that contain non-redundant information since enforcing extraneous constraints is costly and can negatively affect system performance. We conduct two case studies using real business datasets that demonstrate the quality and usefulness of our techniques.