Profiles and contexts axe the main concepts used by modern applications (e.g. e-commerce and recommender systems) to adapt content delivery services to the users' needs, preferences and environment. Although the definitions of the two terms slightly differ from one application to another, there is a general agreement to distinguish them and use them separately or jointly in a given application. When used jointly, the relationship between the two concepts remains often unclear. This paper aims at providing a personalization model that encompasses profile, context, and a formal relationships between the two. This relationship, called contextualization, is represented by a set of ranked mappings, automatically extracted from a usage history (log file of user actions). Profile, context and contextuaUzation constitute three structuring elements over which any personalized system should be built. The proposal is supported by a design platform which helps in instantiating profiles and contexts and in generating contextual mappings between them. An instantiation of the meta model is given for an advanced recommender system, called context-aware recommender system (or CARS for short). This instantiation is followed by an experiment highlighting the benefit of contextualization.