Joint Use of Multiple Learned Statistics for Improving Online Source Selection
Thomas Hernández, Zaiqing Nie, Subbarao Kambhampati · 2004
The autonomous and decentralized nature of available online sources prevents most existing integration systems from supporting flexible query processing that takes into account conflicting user objectives such as coverage, cost-related, or data-quality objectives. To achieve multi-objective query processing, a data integration system must be able to determine which sources are most relevant for a particular query, given the desired objectives. To do so, it must gather and use source-specific statistics. In this paper we present an approach which automatically gathers coverage and overlap statistics as well as response time statistics, and jointly uses these statistics to select relevant sources. We describe our approach and present experimental results done in the context of BibFinder that demonstrate the efficiency and effectiveness of our approach.