3. Source extraction

Data warehouses are fed from multiple sources, autonomous (managed by different, independent systems), heterogeneous (structurally or semantically), possibly unstructured (semi-structured data) or unstructured.

The first step in building a warehouse from such sources is to write an ad hoc program for each one, selecting the relevant data from the source and adapting it to the requirements of the system managing the data warehouse. This approach is particularly demanding, especially when it comes to refreshing the warehouse and adapting to its evolution.

An intermediate approach, designed to accommodate the heterogeneity of source management systems while preserving their autonomy, is to generate an image for each source (the role of the adapter) in a model compatible with that of the warehouse. A generic unification process then feeds the...