The era of "big data" technology has penetrated into all areas of society. In the field of education, personalized teaching has gradually become a new development direction. With the widespread promotion of online education, the adaptive learning system has attracted attention because of its good course recommendation function. As an important part of the general model, the student model reflects the individual characteristics, knowledge status and cognitive ability of the students. The traditional student model judges students based solely on the student’s basic information and simple test scores. This paper proposes a CD-CAT hybrid model based on the actual scene of the learner, retains the original static question bank, and dynamically updates the student model based on the test database of the learner’s different knowledge structure. The knowledge field is divided into multiple categories, and the dynamic allocation of the question bank between categories is carried out. After testing, this dynamic model has a good effect.