The Internet has become an important source of medical information for patients and their family members : search for information about their diseases and recent clinical research, building numeric communities for exchange and sharing of information and of personal experience. However, access to the Internet does not mean access to information. The lack of familiarity with the medical language is a major problem for health consumers in information access and understanding. The aim of this thesis is to analyse and characterize the terms used by nonprofessionals during their discourse on medical topics in order to propose services adapted to their language and to their level of knowledge. Its main production is a health consumer ontology in the breast cancer field, which is based on two types of text corpora : health mediators and health consumers. The elements of this ontology have been analysed on several levels : terms, concepts and relations. The resulting ontology has been the core of a health consumers query reformulation application using spreading activation techniques through a semantic network. The concepts of the ontology represent the nodes in the semantic network. The links between the nodes have weights calculated, either automatically based on the co-occurrence of concepts in a corpus of texts or manually depending on the links type, reflecting the "strength" of the relation between the nodes.
