Overcoming poor data quality: Optimizing validation of precedence relation data

Benedikt Finnah, Jochen Gönsch, Alena Otto

Abstract: Highlights•The effort required to fully validate data is much less than naive estimates.•We propose optimal policies for the problem’s important special cases.•Even straightforward smart policies remove many precedences after several queries.•A new learning-based heuristic shows an even better empirical performance.•In a case study, limited data validation yields double-digit outcome improvement.