Personalization paradigm aims at providing users with the most rel- evant content and services according to many factors such as interest center or location at the querying time. All this knowledge and requirements are orga- nized into user profiles and contexts. A user profile encompasses metadata de- scribing the user whereas a context groups information about the environment of interaction between the user and the system. An interesting problem is there- fore to identify which part of the profile is significant in a given context. This paper proposes a contextualization service which allows defining relationships between user preferences and contexts. Further, we propose an approach for the automatic discovery of these mappings by analyzing user behavior extracted from log files.