Probabilistic Databases

Probabilistic databases are databases where the value of some attributes, or the presence of some records is uncertain, and known only with some probability. Applications in manyareas such as information extraction, RFID and scientific data management, data cleaning,data integration, and financial risk assessment produce large volumes of uncertain data,which are best modeled and processed by a probabilistic database. This book presents the state of the art in representation formalisms and query processingtechniques for probabilistic data. It starts by discussing the basic principles forrepresenting large probabilistic databases, by decomposing them into tuple-independenttables, block-independent-disjoint tables, or U-databases. Then it discusses two classes oftechniques for query evaluation on probabilistic databases. In extensional query evaluation,the entire probabilistic inference can be pushed into the database engine and, therefore,processed as effectively as the evaluation of standard SQL queries. The relational queriesthat can be evaluated this way are called safe queries. In intensional query evaluation, theprobabilistic inference is performed over a propositional formula, called lineage expression:every relational query can be evaluated this way, but the data complexity depends dramaticallyon the query being evaluated, and can be #P-hard. The book also discusses someadvanced topics in probabilistic data management such as top-k query processing, sequentialprobabilistic databases, indexing and materialized views, and Monte Carlo databases.