Canonical structure in attribute based file organization

Abstract

A new file structure for attribute based retrieval is proposed in this paper. It allows queries involving arbitrary Boolean functions of the attribute-value pairs to be processed without taking intersections of lists. The structure is highly dependent on the way in which the file is to be used and is uniquely determined by the specification of the allowed queries. Thus, for example, the structure for retrieval on the basis of ranges of values on a given attribute would be very different from one where only retrieval on the basis of a single value is permitted. The file organization being proposed is based on the atoms of a Boolean algebra generated by the queries. The desirable properties claimed for this structure are proved, and file maintenance questions are discussed.