Article Preview
TopWe can assimilate the problematic of designing a data warehouse to a transformation system which has two main inputs: (1) a dataset usually structured according to a model (for example relational model), which we call the source model, and (2) a set of decision-making needs expressed formally or informally. As output, the system must produce a multidimensional model (star schema, snowflake or facts constellation), which we call the target model. Thus, the role of a multidimensional design method is to specify a set of steps, if possible automated, dealing with the inputs to produce the target model.
Several works have been proposed in the context of multidimensional design methods, which can be classified into three categories (Romero & Abelló, 2009):
The main criterion for this classification -cited in the majority of the papers in the field-, is how they deal with the two inputs of the transformation system. We present below these three categories, focusing on a few representative methods for each category, in order to point the basic fundamentals of each category. Then, we will discuss their advantages and drawbacks.