Mirjana Mazuran: Semi-automatic support for evolving tuple constraints and functional dependencies

Abstract: During the life of a database, systematic and frequent violations of a given constraint may suggest that the represented reality is changing and thus the constraint should evolve with it. We focus on tuple constraints and functional dependencies and propose a method to (i) identify the violated constraints and (ii) support their evolution when it is necessary to update them. To repair tuple constraints we trace all the violations that were refused by the DBMS and use frequent itemset mining to find those itemsets that represent significant violations. Then we use these itemsets to propose a modified version of the initial constraint to the designer. Instead, for functional dependencies, our method relies on the use of (i) confidence, as a measure that is associated with each dependency and allows us to understand "how far" the dependency is from correctly describing the current data and (ii) goodness, as a measure of balance between the data satisfying the antecedent of the dependency and those satisfying its consequent. Our current line of research focuses on integrating these two strategies for constraint evolution into a single framework for constraint violation monitoring.