In recent years, an increasing interest in recommendation systems has emerged both from the research and the application point of view and in both academic and commercial domains. The majority of comparison techniques used for formulating recommendations are based on set-operations over user-supplied terms or internal product computations on vectors encoding user preferences. In both cases however, the “identical-ness” of terms is examined rather than their actual semantic relevance. This paper proposes a recommendation algorithm that is based on the maintenance of user profiles and their dynamic adjustment according to the users’ behavior. Moreover, this algorithm relies on the dynamic management of communities, which contain “similar” and “relevant” users and which are created according to a classification algorithm. The algorithm is implemented on top of a community management mechanism. The comparison mechanism used in the context of this work is based on semantic relevance between terms, which is evaluated with the use of a glossary of terms.