Today, there is a growing need for a user to able to express and get answers to more complex requests, including multiple functionalities, conditions, constraints and objectives. Complex requests including multiple functionalities cannot usually be answered with one single Web service. As multiple services are needed, the problem is then to find good combinations using the available services but also to select the best ones according to the user constraints and to the objectives expressed in the request. This thesis contributes to answering this issue. It focuses on the problem of semantic Web services composition and optimization to answer such requests.

The problem requires formalizing: request, services and composition. While Web services are described semantically using the W3C OWL-S ontology, we design a compatible ontology for the request semantic description, which we name OWL-CR.

For the automatic design of service composition satisfying a request, we propose a new model for the representation of semantic Web services composition and an algorithm that builds compositions based on this model.

For composition optimization, we experiment an algorithm called Martins' algorithm, which provides good results to select best compositions. Genetic Algorithms can be used in the case of global dependencies, when the composition model is a cluster set. In this case, the set of solutions cannot be defined a priori and the application of approximated optimization methods is necessary.