Preference-driven querying of inconsistent relational databases

One of the goals of cleaning an inconsistent database is to remove conflicts between tuples. Typically, the user specifies how the conflicts should be resolved. Sometimes this specification is incomplete, and the cleaned database may still be inconsistent. At the same time, data cleaning is a rather drastic approach to conflict resolution: It removes tuples from the database, which may lead to information loss and inaccurate query answers. We investigate an approach which constitutes an alternative to data cleaning. The approach incorporates preference-driven conflict resolution into query answering. The database is not changed. These goals are achieved by augmenting the framework of consistent query answers through various notions of preferred repair. We axiomatize desirable properties of preferred repair families and propose different notions of repair optimality. Finally, we investigate the computational complexity implications of introducing preferences into the computation of consistent query answers.