Scrubbing query results from probabilistic databases

Abstract

Queries over probabilistic databases lead to probabilistic results. As the process of arriving at these results is based on underlying data probabilities, we believe involving a user in the loop of query processing and leveraging the user's personal knowledge to deal with uncertain data, will enable the system to scrub (correct) and tailor its probabilistic query results towards a better quality from the perspective of the specific user. In this paper, we propose to open the black box of a probabilistic database query engine, and explain to the user how the engine comes up with the probabilistic query result as well as which uncertain tuples in the database the result is derived from. In this way, the user based on his/her knowledge about uncertain information can not only decide how much confidence to be placed on the query engine, but also help clarify some uncertain information so that the query engine can re-generate an improved query result. Two particular issues associated with such a probabilistic database query framework are addressed: (i) how to interact with a user for answer explanation and uncertainty clarification without bringing much burden to the user, and (ii) how to scrub/correct the query result without incurring much computation overhead to the query engine. Our performance study demonstrates the accuracy effectiveness and computational efficiency achieved by the proposed framework.