Visualisations novatrices pour la comprehension de reseaux et de logiciels complexes

The presentation of information can potentially benefit from our biological visual system, which evolved over hundreds of millions of centuries, to facilitate the discovery of hidden secrets, new patterns or unexpected relationships in data. There is much information in the world, of various kinds, structured or unstructured, that we seek to understand from different points of view. In the literature, visualization techniques can be useful in specific contexts, but each has its own limitations. However, there are still many opportunities to create and explore new ways of representing data and networks. The validation of these novel techniques remains a challenge. Indeed, these new interfaces must be easy to use, but at the same time, help users analyze and explore complex data.;In particular, important data can be modeled as networks, such as social interactions or financial transactions, but it is harder to visualize these networks if they are large, include hierarchies and/or evolve over time. This thesis focuses on designing new visualization techniques of networks to facilitate the understanding of data.;In the goal of comparing alternative visualization techniques to support data comprehension tasks, we have classified previous work using taxonomies. We were thus able to reveal new interesting combinations (or hybrids) to visualize static and dynamic networks.;Our contributions cover different aspects of the visualization of complex and dynamic graphs. The first chapter focuses on visualizing static compound graphs by combining existing techniques. The prototype that we describe in the second chapter also explores combinations of representations, but can be used to show dynamic graphs. Finally, the third chapter present a new visual and practical approach to track complex software structures in evolving software designs (modeled as networks).;In our first prototype (TreeMatrix), parts of a network are shown using matrices and nodelink diagrams, while tree structures are shown using icicle diagrams and nested enclosures. One benefit of this technique is that it helps visualize dense networks, without affecting the ability to comprehend higher level relationships. A user study has shown advantages of using our novel technique to discover and organize links in a software design, compared to a mature commercial tool (Lattix).;We also combined techniques in novel ways in our second prototype (DiffAni) to visualize evolving networks. DiffAni is the first hybrid approach that can be used for dynamic networks, and its validation using human subjects revealed benefits in some cases. Thus, animation should be used sparingly, i.e., when there is significant movement in a network. These results along with our taxonomies could guide the exploration of more novel combinations in future work.;Finally, our third prototype (IHVis) focused on exploring and tracking evolving software design structures. Using data collected in software source code repositories, our new visual tool was able to discover interesting cases of stability protection, and refactorings. Also, several human subjects reported their findings, such as the implementation of new variations in features.;In summary, this thesis present novel and useful ways to show complex and evolving network structures. Our main contributions are (1) the exploration of design spaces for novel network visualizations using taxonomies, (2) implementation of prototypes combining techniques to visualize compound graphs and dynamic networks, (3) a new visual method to explore extensions and instabilities in evolving software, (4) the evaluation of these techniques in user studies.