Extracting symbolic objects from relational databases

Our aim is to define operators to retrieve groups of individuals from a relational database. The way we describe these groups makes it possible to analyse them by symbolic data analysis methods which extend classical ones to more complex data. In so far as the input consists of groups of data extensionally defined in the database, our problem is to find the best description representing each group in the formalism (called symbolic object) of symbolic data analysis. In our process, we take into account data from tables together with additional knowledge such as taxonomies. To describe each group, we perform a generalization step and a specialization one. Final descriptions are based on the notion of homogeneity within a group and they minimize a volume criterion.