Abstract
The modelization of the Web graph and the modelization and the extraction of communities in the graph of the Web are the subject of this thesis, which is divided into two parts. The first part makes an analysis of large graphs and introduced a new model of random crawls. We starts by defining the common properties of networks, then gives some random models for the generation of networks. To finish, we proposes a new model of random crawls.<br />Then, the second part proposes two models of emergence of community in the networks. After a remainder on the algorithms of classification: PageRank and HITS is presented the gravitational model in which the nodes of a network are mobile and interact to the links between them. The communities emerge quickly after some iterations. The second model is an improvement of the first, the nodes have now an objective which consists in reaching their communities.