Attribute metadata for relational OLAP and data mining

Abstract

To build the d-dimensional datacube, for on-line analytical processing, in the relational algebra, the database programming language must support a loop of d steps. Each step of the loop involves a different attribute of the data relation being cubed, so the language must support attribute metadata. A set of attribute names is a relation on the new data type, attribute. It can be used in projection lists and in other syntactical postions requiring sets of attributes. It can also be used in nested relations, and the transpose operator is a handy way to create such nested metadata. Nested relations of attribute names enable us to build decision trees for classification data mining. This paper uses OLAP and data mining to illustrate the advantages for the relational algebra of adding the metadata type attribute and the transpose operator.