A formal treatment of imperfect information in database management

Traditionally, database management systems have been presented and evaluated with the assumption of perfect information. In this thesis, the problem of dealing with imperfection in a database is discussed and a formal framework for a general treatment of imperfection is proposed. Incompleteness and inconsistencies in the stored values are two manifestations of imperfection we consider. In order to have a clear, well-defined framework and to illustrate our concepts with specific examples, we concentrate on one database model--the relational model. The organization of databases, the semantics of operations on them (queries and modifications) and certain semantic rules (functional dependencies) are examined in detail when imperfect information is present. Finally, it is shown that our work has an important application to traditional relational database theory, where a basic and controversial assumption is weakened with the use and careful treatment of imperfection.