In this paper we claim that learning in a first order logic framework is crucial for the future of user modeling applications in the context of the Semantic Web (SW in the remaining of the paper). We first present some works that have currently been done for designing first order logic based languages, for reasoning in the SW. In the context of user modeling in the SW, it would then be relevant to use such languages to model user’s behaviors and preferences. We show that discovering knowledge in the context of such languages could be done using multi-relational data mining that has already provided efficient prototypes. Nevertheless, some work remains to be done in order to use them in that context and we give some directions for that purpose.