Model-Driven Data Warehouse Automation: A Dependent-Concept Learning Approach

1. Introduction

The decision support systems and business intelligence systems (Turban et al., 2010; Poe et al., 1997) are the areas of the information systems discipline that is focused on supporting and improving decision-making across the enterprise. The decision-making process is a strategic asset that helps companies to differentiate themselves from competitors, improve service, and optimize performance results. The data warehouse (Kimball and Ross, 2002, 2010) is the central component of current decision support and business intelligence systems and is responsible for collecting and storing useful information to improve decision making process in organization. Several data warehouse design frameworks and engineering processes have been proposed during the last few years. However, the framework-oriented approaches (Luján-Mora et al., 2006; Prat et al., 2006; Simitsis, 2005) fail to provide an integrated and a standard framework that is designed for all layers of the data warehousing architecture. The process-oriented approaches (Westerman, 2001; List et al., 2000; Kaldeich and Sá, 2004) fail, also, to define an engineering process that handles the whole development cycle of data warehouse with an iterative and incremental manner while considering both the business and the technical requirements. In addition, not much effort was devoted to unify the framework and the process into a single integrated approach. Moreover, no intelligent and automatic data warehouse engineering method is provided.

The model-driven data warehouse gathers approaches that align the development of the data warehouse with a general model-driven engineering paradigm (Bézivin, 2006). The model-driven engineering is mainly based on models, meta-models, and transformation design. Indeed, model-driven strategy encourages the use of models as a central element of development. The models are conforming to metamodels and the transformation rules are applied to refine them. Therefore, transformations are the central components of the each model-driven process. However, transformation development is a very hard task that makes the model-driven approach more complex and entails additional costs. So, designers or programmers must have high skills in the corresponding metamodels and the transformation languages as the query-view-transformation (Object Management Group/QVT, 2010). In addition, data warehousing projects require more knowledge about the underlying business domain and requirements. This raises many risks and challenges during the transformations design. One of the main challenges is to automatically learn these transformations from existing project traces. In this context, the model transformation by-example (introduced by Varró, 2006a) is an active research area in model-driven software engineering that uses artificial intelligence techniques and proposes to automatically derive transformation rules. It provides assistance to designers in order to simplify the development of model transformations and it reduces complexity, costs and time of development.