Generally, the user requires customized information reflecting his current needs and interests that are stored in his profile. Therefore, taking into account this latter is significant to improve the recommended results. Along the day, the user is more and more active. Consequently, his interests evolve over time at the moment of updating his profile and they are distributed over different systems. This updating was the subject of several research works, in order to reduce the overload of the user’s profile, either by using the machine learning techniques or the notion of temperature. In this paper, we propose an approach to improve the updating of user’s profile by respectively merging the temperature and the k-means leaning algorithm.