Pervasive computing calls for more and more re-engineering on Web sites. Nowadays, Web sites are often designed to be displayed on a personal computer. However, alternative versions are proposed more and more often. Zoom, focus, and thumbnail systems make the browsing easier but are not generally applicable to other devices. In our opinion, it is necessary to be able to adapt a Web document according to the constraints of the browsing device. This thesis proves that a meta-heuristic approach allows to make the re-engineering of a Web site automatic for a device whose charactéristics are not known in advance. In order to do so, we propose an architecture coupled with a module able to provide browsing session in the case where the Web document has to be split in chuncks. Device characteristics being not known in advance, we also introduce a system able to gather passively the device information. The work relies on an adapting model inspired by the transcoding. Our model appeals to techniques called transformations. Transformations are multiple, so it is necessary to determin, for each context, which ones must be applied and in which order. This sequencing of transformations is named composition. A genetic algorithm allows us to find an approximate solution for the problem of the best composition. Finally, the thesis presents the implemetation of this model, the results of our experimentations and proposes an interpretation of the performance of the system
