A HYBRID SEMI AUTOMATIC APPROACH FOR THE DESIGN OF DATAWAREHOUSE CONCEPTUAL MODEL

Abstract

Organizations strive to build data warehouses (DW) to support their decision making process. The studies show that the most important task for organization’s manager is to take the right decision in the right time. Hence, the decision-making process is considered among the main important goals to be achieved by competitor enterprises. The design process of a DW raises many problems and is considered as a complex and tedious task. DW designers follow different levels concerning the design process. Among these levels, the conceptual level received a significant coverage in the literature. There are many directions one can follow to design a DW. Indeed, some designers prefer to start from user requirements (User requirements driven or top-down approaches), another camp of DW designers prefers to start the design process from the data source (DS) data-model (data-source driven or bottom-up approaches). Recently, DW designers follow a hybrid approach that benefits from the features of the above two approaches. Putting these two sources together raises a structural problem known as heterogeneity; using ontology while designing a DW helps organizations to solve the problems of semantic heterogeneity between data sources and users’ requirements. This thesis uses a hybrid semi-automatic approach relying on a semantic resource for the design of a DW conceptual model. First, a heuristic-rule based approach to generate star schemas from relational data source has been defined. To come up with this task, the tables’ structures of the relational DS from the DBMS repository have been extracted; as well, a reverse engineering process to classify these DS tables into tables modeling entities and tables modeling relationships has been conducted. This classification helps to generate a multidimensional model (star schema) from the DS data-model. Second, a natural language approach to build star schemas from business requirements has been employed. This task encompasses three steps: business requirements elicitation, user requirements normalization and multidimensional star schemas generation. Third, a matching process between star schemas from the DS and that from business requirements is tackled. The matching process is relying on a semantic resource, WordNet in particular, to overcome heterogeneity; as well, the DW designer has been allowed to intervene to approve the star schemas. The approach has been tested on various examples from the literature. The generated results show the feasibility of the proposed approach for generating approved star schemas.