Fonctions d'oubli et résumés dans les entrepôts de données

The amount of data stored in data warehouses grows very quickly so that they get saturated. To overcome this problem, the solution is generally to archive older data when new data arrive if there is no space left. This solution is not satisfactory because data mining analyses based on long term historical data become impossible. As a matter of fact data mining analysis cannot be done on archived data without re-loading them in the data warehouse; and the cost of loading back a large dataset of archived data is too high to be operated just for one analysis. So, archived data must be considered as lost data regarding to data mining applications. In this thesis, we propose a solution for solving this problem: a language is defined to specify forgetting functions on older data. The specifications include the definition of some summaries of deleted data to define what data should be present in the data warehouse at each step of time. These summaries are aggregates and samples of deleted data and will be kept in the data warehouse. The goal of these forgetting functions is to control the size of the data warehouse. This control is provided both for the aggregate summaries and the samples. The specification language for forgetting function is defined in the context of relational databases. Once forgetting functions have been specified, the data warehouse is automatically updated in order to follow the specifications. This thesis presents both the language for specifications, the structure of the summaries, the algorithms to update the data warehouse and the possibility of performing interesting analyses of historical data.