This paper deals with the automatic building of personalized hypermedia. We build upon ideas developed for educational hypermedia. A standard way to build adaptive educational hypermedia relies on the definition of a domain model and the use of overlay user models. Since much work has been done on learning user models and adapting hypermedia based on such user models, the core problem lies in the automatic definition of a domain model for a static hypermedia. We describe an approach to automatically learn from the hypermedia content such a domain model. This model is a concept hierarchy where concepts are identified by sets of keywords learned from the collection. We propose the use of visualization techniques such as treemaps in order to monitor and analyze efficiently user and domain models.