Design And Implementation Of Modeling Students Based On Bayesian Networks

First of all, the states-of-the-art on distance learning system is briefly introduced in this dissertation. Among all of correlative technologies, the personalized and adaptive learning system is really a hot topic and also quite difficult to be solved. Furthermore, the implementation of such a learning system depends directly on modeling student.As aforementioned, modeling student is a crucial computer technology on the relevant research area. Based on the constructionism and the cognitive flexibility theory, an overlay student model is established by the aid of Bayesian network approach. The model designed proves to be efficient and feasible to realize the adaptability and the individuation of the learning systems. During the procedure of modeling student, student information is divided into the field-specific one and the field-nonspecific one. To model the field-specific information is to convert the curricula into its version on the Bayesian networks. And to model the field-nonspecific information is to evaluate the accepted level of the new knowledge for students. "Java Tutorial" is taken as an example curriculum, which is disassembled into 298 knowledge items and the mastery of each knowledge item can be expressed into 4 levels. Thereafter, partial order among these knowledge items is set up so as to ascertain the causality on Bayesian networks. Then the values of transcendent probabilities for all of the knowledge items are assigned and an overlayBayesian student model can be formed.However, reasoning on student model is a NP hard question in fact, since undirected loops are embedded in the Bayesian student model. To solve this problem, a revised clustering algorithm is enhanced by means of the joint tree during reasoning process. More specific, the student model is transformed into a junction tree, by means of Moral Graph and triangulation. Therefore, when the student information is returned, the model can update all the item distribution probabilities with the help of self-study ability of the Bayesian networks. Finally, accurate testing of the mastery of new knowledge is fulfilled, as a result, the teaching and learning procedure is individuated.