Sensitivity Analysis of Answer Ordering from Probabilistic Databases.

Queries over probabilistic databases result in probabilistic answers, which are often ranked according to certain ranking criteria. As the probabilities of the basic tuples may be imprecise and erroneous, and their perturbations may lead to great changes in answer ordering, sensitivity analysis like "which basic input probability change can substantially alter the ranked result?", "which basic probability change will make a certain element top-ranked?", "which basic probability change will swap the positions of the firstly and secondly ranked elements?" thus arise. The sensitivity analysis of top-K probabilistic query has been touched in the literature, mainly concerning the change of the answer list as a set. However, the ordering of the elements in the answer list matters highly for certain applications. In this paper, we categorize a variety of such kinds of ordering sensitivity questions into list-oriented or element-oriented, and formulate the sensitivity analysis problem for answer ordering returned from probabilistic top-K queries and probabilistic top-K aggregation queries. We develop a modular approach to quantitatively compute sensitivity of answer ordering, where four basic processing modules are identified. Optimization strategies are also presented for performance improvement. Experimental results on both synthetic and real data demonstrate the effectiveness and efficiency of the proposed solutions. © 2013 Springer-Verlag.