Query processing techniques in probabilistic databases

Abstract

Probabilistic databases have many diverse applications due to inherent uncertainty in those applications. In this paper, we review how uncertainty is managed in probabilistic databases. Two techniques of query evaluation on probabilistic databases exist, namely extensional and intensional. Extensional query evaluation directly operates on tuple probability and thus it is quiet efficient. But, for some queries it gives wrong results. We have to find safe query plan if one exists for a given query. Intensional query evaluation derives probability inference using lineage expression. Intensional semantics can become impractical for complex queries as data complexity depends on the query and the instance. We discuss various queries including joins, top-k, skyline, aggregates on probabilistic databases. In addition, we provide outline of most popular probabilistic databases: MayBMS and Trio. As commercial probabilistic database management system is still lacking, we hope this review will motivate database researchers to develop concrete probabilistic database management system.