A system prototype for computing probabilistic query answers over inconsistent databases

Several techniques for repairing and querying inconsistent databases have been proposed in recent years. In most of them, the repair strategy consists in performing minimal sets of tuple insertions and deletions. The consistent query answers are those query answers which can be derived from every consistent repaired database. A problem with such techniques is that, often, useful information contained in the source database is not preserved in the repaired ones. Another problem is that in the computation of query answers we are not able to discriminate among "undefined" tuples, namely tuples which can be derived from a non empty proper subset of the repaired databases. In order to cope with the aforementioned problems, recently, there have been proposals for repairing databases by means of tuple updates and computing probabilistic query answers. In this paper we discuss a probabilistic approach to compute query answers and present a system prototype. More specifically, we consider databases which can be inconsistent with respect to canonical sets of functional dependencies. Given a Relational Algebra (RA) query to be evaluated over a possibly inconsistent database, the technique firstly computes a probabilistic database, where each tuple is associated with a probabilistic event, and next evaluates the query (which uses the probabilistic RA instead of standard RA) on the probabilistic database. The probability associated with a tuple in the answer gives a measure of the number of repaired databases from which the tuple can be derived. A system prototype for computing probabilistic query answers is presented.