A framework for multidimensional design of data warehouses from ontologies

Abstract

Some research efforts have proposed the automation of the data warehouse design in order to free this task of being (completely) performed by an expert and facilitate the whole process. Most advanced approaches exclusively work over relational sources and perform a detailed analysis of the data sources to identify the multidimensional concepts in a reengineering process. Starting from a logical schema, however, may present some inconveniences. A logical schema is tied to the design decisions made when devising the system and these decisions either made to fulfill the system requirements (for instance, improve query answering, avoid insertion / deletion anomalies, preserve features inherited from legacy systems, etc.) or naively made by nonexpert users, have a big impact on the quality of the multidimensional schemas got by current automatable approaches.In this paper, we introduce our approach for automatically deriving the multidimensional schema from a domain ontology. Our goals are mainly two: i) we want to improve the quality of the output got (by working over a conceptual formalization of the domain instead of a logical one) and ii) we want to automate the process. This second goal is the main reason for choosing ontologies instead of other conceptual formalizations, as ontology languages provide reasoning services that will facilitate the automation of our task.

A Comprehensive Framework on Multidimensional Modeling

“…Clearly, this is the major flaw of current approaches, which do not suit well for the wide range of real projects a designer could meet. Interestingly, it has already been pointed out [18] that, given a specific design scenario, the necessity to provide requirements beforehand is smoothed by the fact of having semantically rich data sources. In lack of that, requirements gain relevance to extract the MD knowledge from the sources.…”

“…This prerequisite is not easy to fulfill in many real organizations and in order to solve this problem, current automatable methods directly work over relational databases (i.e., getting up-to-date data). To our knowledge, only three exceptions exist to this rule [20,19,13], which automate the process from ER schemas (the first one) and ontologies (the other two). Consequently, all these methods (or stages within hybrid approaches) follow a supply-driven paradigm and thus, rely on a thorough analysis of the sources.…”

“…Eventually, they put the burden of (manually) analyzing and filtering results provided onto the designer's shoulder, but the time-consuming nature of this task can render it unfeasible when large data sources are considered. To our knowledge, only [19] filters out results obtained prior to show the results to the user. Furthermore, all these facts directly affect the computational complexity of SDAs.…”

An Ontology based Hybrid Approach to Derive Multidimensional Schema for Data Warehouse

“…We use certain general parameters and parameters specific to multidimensional element identification for comparison. Table 2 represents the comparison of the proposed approach with AMDO [17] and AMDMM [18]. The AMDMM approach follows a hybrid methodology similar to the proposed one by analyzing the source and requirements before the design task.…”

“…Here only if the requirements are stated clearly the design can be successful. In [17] AMDO (Automating Multidimensional Design from Ontologies) they use three criteria such as multidimensionality, the multidimensional space arrangement constraint and the summarization integrity constraint in order to carry out the design task from the source ontology. They use basic and generic reasoning algorithms to automatically derive facts and dimensions.…”