Learner modalization in adaptive systems contains several indicators, such as domain knowledge, learning performance, goals, tasks, background, learning styles and learning environment. Even if there are several methods for initializing the learner model, such as the stereotype model, or learner profiles, these models does not manage the side of uncertainty in the dynamic modeling of the learner. The main hypothesis of this article is the initialization of the learner model based on the combination of the Bayesian networks and the stereotypes methods. To achieve this objective, it is necessary to ask why and how to initialize a model of the learner by combining the method of stereotypes with Bayesian networks? What steps can be taken to move from the learner information gathering phase to the initialization of a learner model in a comprehensive way? We focus in this article on the first two steps in the process of adaptation, collecting data about the user, and initiating the learner model. In order to carry out a complete initialization of this model, a combination of the stereotypes method to process the content of the specific domain of information, and the Bayesian networks to process the contents of the independent domain of information have been used.