Mixed Driven Refinement Design of Multidimensional Models based on Agglomerative Hierarchical Clustering

Abstract

Abstract:Data warehouses (DW) and OLAP systems are business intelligence technologies allowing the on-line analysis of huge volume of data according to users' needs. The success of DW projects essentially depends on the design phase where functional requirements meet data sources (mixed design methodology) (Phipps and Davis, 2002). However, when dealing with complex applications existing design methodologies seem inefficient since decision-makers define functional requirements that cannot be deduced from data sources (data driven approach) and/or they have not sufficient application domain knowledge (user driven approach) (Sautot et al., 2014b). Therefore, in this paper we propose a new mixed refinement design methodology where the classical data-driven approach is enhanced with data mining to create new dimensions hierarchies. A tool implementing our approach is also presented to validate our theoretical proposal.

A Semi-Automatic Design Methodology for (Big) Data Warehouse Transforming Facts into Dimensions

“…The new hierarchies can be used to analyze other facts in a constellation schema. This work extends that proposed in [5] by using several facts for the creation of hierarchies. 2) Reduction: The previous methodology created several hierarchies that are added to the new dimension(s).…”

“…In this work, our aim was thus to extend a previous study [5] to provide a semi-automatic mixeddriven design methodology that creates dimension hierarchies originating from factual data in a constellation schema. The main contributions of this paper are: 1) Transformation: A formal approach is proposed to automatically calculate dimension hierarchies using factual data.…”