Evaluating Probabilistic Queries over Uncertain Matching

Abstract

A matching between two database schemas, generated by machine learning techniques (e.g., COMA++), is often uncertain. Handling the uncertainty of schema matching has recently raised a lot of research interest, because the quality of applications rely on the matching result. We study query evaluation over an inexact schema matching, which is represented as a set of ``possible mappings'', as well as the probabilities that they are correct. Since the number of possible mappings can be large, evaluating queries through these mappings can be expensive. By observing the fact that the possible mappings between two schemas often exhibit a high degree of overlap, we develop two efficient solutions. We also present a fast algorithm to compute answers with the k highest probabilities. An extensive evaluation on real schemas shows that our approaches improve the query performance by almost an order of magnitude.