Managing and processing large amounts of heterogeneous data often requires mapping a structure to the data. This structure might be the result of an analysis of the relations existing between the data without considering their generative processes (data mining). In another hand, one may also want to associate meta-data to the original data so as to qualify them with higher level information (knowledge representation). Meta-data are then added to the original data. Structuring the data is often mandatory so as to use them in compex systems (e.g. learning algorithms, task adaptation). Moreover, high-level data representation can lead to the application of reasoning techniques to the systems having generated the data (symbokic learning).