Abstract
The thesis is focused on the features of real interaction networks, especially on the “small world” and “scale free” properties. It presents various filtering, clustering and visualization techniques of these networks.<br />The “scale free” property comes from these interaction networks being most often in growth process: the new nodes tend to connect to high degree nodes. This process is called “preferential attachment”. The “small world” property recall the saying: “my friends' friends are my friends”.<br />Interaction networks usually have a dense core that is difficult to analyse and to visualize with classical clustering and drawing techniques. This study develops a new filtering technique allowing extracting a “tree like” structure also having “small world” and “scale free” properties. The resulting network is organized into a tree of nested silhouettes. This multi scaling structure is easily drawn and explored. It presents a contextual organization around user focus. The new clustering technique optimizes the quality criteria listed and presented in the thesis. Moreover, the silhouettes'content interpretation emphasizes the quality of the joint use of filtering and<br />clustering techniques.