Imprecise information and uncertainty in information systems

Reviewer: Karen Sparck-Jones

Awkward database queries are those for which values are imprecise (with respect to domain or range) or unknown, or attributes are inapplicable. Being able to deal satisfactorily with data distinctions of this kind is important for real applications. Morrissey claims that this approach is an advance on earlier work in offering both an appropriate and coherent semantics for query evaluation covering the different cases and an operational measure of uncertainty for ordering retrieved data that do not precisely satisfy the query. The form of data representation and method of query evaluation for different simple and complex types of query are described in detail, showing how first precise and then imprecise data sets are retrieved. The members of the second set can in turn be ranked by their degree of uncertainty, using information-theoretic measures based on either self-information or entropy. The paper is clearly written and well illustrated by examples. The way the different kinds of data are retrieved looks convincing. Unfortunately the treatment of uncertainty, though implemented for a prototype office information system, has not been seriously tested. Such testing is clearly essential, as approaches of the kind presented all tend to look plausible but need to be subjected to the hard reality of actual hairy data to demonstrate their solidity. The paper is nevertheless an interesting start and, as it gives a clear and detailed account of the author's work, a useful introduction to the area.