One of the key issues in processing massive and heterogeneous data is to understand them in order to be able to extract new knowledge or verify a hypothesis put forward by an expert in the field. In this framework, we are interested in the unsupervised classification of data (particularly in relational data) and in the definition of metrics to quantify the importance/influence of elements and the relationships between them. We are also interested in supervised learning for automatic evaluation of the effectiveness of data visualisation techniques. In addition, we are developing visualization tools to facilitate the understanding of neural networks.