Typically, personalized information recommendation services automatically infer a user profile, a structured model of the user interests, from documents the user already deemed as relevant. Traditional keyword-based approaches are unable to capture the semantics of the user interests. This work proposes a strategy consisting of two steps. The first one is a semantic indexing procedure based on a word sense disambiguation strategy which exploits the WordNet lexical database to select, among all the possible meanings (senses) of a polysemous word, the correct one. In the second step, semantically indexed documents are mined by a naive Bayes learning algorithm that infer semantic, sense-based user profiles. Two experimental sessions were carried out to compare the performance of keyword-based profiles to that of sense-based profiles. We measured both the classification accuracy and the effectiveness of the ranking imposed by the two different kinds of profile on the documents to be recommended. The main outcome of both experiments is that the classification accuracy is improved without improving the ranking. Personalized systems adapt their behavior to individual users by learning their preferences during the interaction in order to construct a user profile that can be later exploited in the search process. Traditional keyword-based approaches are primarily driven by a string-matching operation: If a string, or some morphological variant, is found in both the profile and the document, a match is made and the document is considered relevant. String matching suffers from problems of polysemy, the presence of multiple meanings for one word, and synonymy, multiple words having the same meaning. The result is that, due to synonymy, relevant information can be missed if the profile does not contain the exact