Nowadays many newspapers and news agencies offer personalized information access services and, moreover, there is a growing interest in the improvement of these services. In this paper we present a methodology useful to improve the intelligent personalization of news services and the way it has been applied to a Spanish relevant newspaper: ABC. Our methodology integrates textual content analysis tasks and machine learning techniques to achieve an elaborated user model, which represents separately short-term needs and long-term multi-topic interests. The characterization of user’s interests includes his preferences about structure (newspaper sections), content and information delivery. A wide coverage and non-specific-domain classification of topics and a personal set of keywords allows the user to define his preferences about content. Machine learning techniques are used to obtain an initial representation of each category of the topic classification, . Finally, we introduce some details about the Mercurio system, which is being used to implement this methodology to ABC. We describe our concrete experience and an evaluation of the system in comparison with other commercial systems.