\input zb-basic
\input zb-ioport
\iteman{io-port 05976389}
\itemau{Suciu, Dan; Olteanu, Dan; R\'e, Christopher; Koch, Christoph}
\itemti{Probabilistic databases.}
\itemso{Synthesis Lectures on Data Management 16. San Rafael, CA: Morgan \& Claypool Publishers (ISBN 978-1-60845-680-2/pbk; 978-1-60845-681-9/ebook). xiv, 164~p. (2011).}
\itemab
Probabilistic databases represent a new approach to data modeling in that they are databases where the values of some attributes or the presence of some tuples are uncertain and known only with a certain probability. This uncertainty view has numerous applications (e.g., in data integration, data cleansing, scientific databases, RFID data management, financial risk assessment), yet requires not only an appropriate representation formalism, but also an in-depth revision of basic database concepts such as query processing or indexing. The book casts probabilistic databases in terms of the relational model of data, and puts an emphasis on queries and query evaluation under the constraint that attribute values no longer represent assured facts of some real-world application. It concisely introduces key concepts (such as types of uncertainty, types of probabilistic databases, lineage, or safe queries). Probabilistic databases are classified into tuple-independent databases, where tuples represent independent probabilistic events, block independent-disjoint databases, where tuples are partitioned into blocks such that all tuples within a block are disjoint events and all tuples from different blocks are independent events. Such decomposition is achieved by a process resembling traditional database normalization. Since an instance of a probabilistic database can be one of several states (has several ``possible worlds'', each to be assumed with a certain probability), the semantics of a query needs a careful definition: A query is applied to all possible worlds, but its result can either consist of all possible answers (where each is a set of tuples), or it consists of a single set of tuples that contains all possible answers combined. The book discusses two classes of query evaluation techniques in detail: extensional and intensional query evaluation, which exhibit the classical ``expressiveness vs. complexity'' trade-off of relational query evaluation. The book is written by specialists for specialists; in its 160 pages, it gives a state-of-the-art account of probabilistic databases and the research that has been done in this field to date, with an excellent account of the relevant research literature. It not only surveys the main concepts of probabilistic databases, but also introduces various advanced topics, such as Monte Carlo databases or materialized views. It does so using examples wherever possible, but also in a strict formal manner. In conclusion, it is a very valuable addition to the database literature. For readers it is advisable to have a background both in database theory and in probabilistic inference, which makes the book suitable for advanced graduate classes or established researchers interested in the topic.
\itemrv{Gottfried Vossen (M\"unster)}
\itemcc{}
\itemut{database; probabilistic database; query processing; extensional query; intensional query; query evaluation}
\itemli{doi:10.2200/S00362ED1V01Y201105DTM016}
\end