The recent evolution of e-commerce emphasized the need for more and more receptive services to the unique and individual requests of users. Personalization became an important business strategy in Business to Consumer commerce, where a user explicitly wants the e-commerce site to consider her own information such as preferences in order to improve access to relevant products. In this work, we present a personalization component that uses supervised machine learning to induce a classifier able to discriminate between interesting and uninteresting items for the user. The prototype system exploits a content-based technique, which makes use of textual annotations usually describing the products offered by e-commerce sites. Experimental results demonstrate the effectiveness of the method and encourage the integration of the prototype in the personalization module developed in the COGITO project, which aims at improving consumersupplier relationships in future e-commerce using advanced technologies.