Open Archive Toulouse Archive Ouverte (oatao) Multidimensional Database Modelling with Differentiated Multiple Aggregations

OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. Multidimensional database modelling with differentiated multiple aggregations. Many solutions have been defined for multidimensional database modelling. These propositions consider the same aggregation function to determine the values of an indicator according to different levels of granularity into the multidimensional space. We provide a more flexible conceptual model that supports multiple differentiated aggregations. Multiple aggregations allow associating different aggregation functions to the same measure for each dimension and for each hierarchy. Differentiated aggregation allows specific aggregations at each level (parameter). Our model is based on a double graphical formalism, expressive enough to control the validity of aggregation functions. We also study the consequences of this conceptual modelling for building lattices of pre-computed aggregates in a relational online analytical processing (R-OLAP) environment. 1. Introduction Online analytical processing (OLAP) analyses consist in monitoring indicators represented as points in a multidimensional space using different analysis axes. According to hierarchies associated with analysis axes, aggregation functions are used for obtaining a synthetic view of indicator values. Data are grouped according to the selected detail level and aggregated by the functions. Drilling operations (roll-up and drill-down), often used in OLAP analyses, make intensive use of these aggregation functions. Multidimensional databases (MDB) provide an appropriate framework for decision-maker analyses. However, in some cases, the default aggregation functions provided may be maladjusted for decision-making. For example, the analysis of temperatures can be performed uniformly with a same aggregation function. The annual maximum temperature is obtained from the monthly maximum temperatures. The latter are obtained from daily maximum temperatures. However, the analysis of temperatures by departments is usually based on the average temperatures by cities, while the analysis of temperatures by region can be performed from an aggregation of the department temperatures, which takes into consideration the department's area (a weighted average). Thus, this analysis involves non-uniform different aggregation functions (such as average or weighted average) according to a considered analysis level (department,