With the development of Intelligent Teaching Systems (ITS), the importance of student model has gradually increased. Knowledge tracking (KT), as a dynamic cognitive diagnostic method, is widely used to predict students’ future performance based on their historical performance. Existing studies have shown that probabilistic graphical models are favored for their interpretability, scalability, and good predictive performance. However, these models generally lack modeling of students’ abilities, which limits their effectiveness in intelligent teaching systems. Thus, this paper proposes a Bayesian Knowledge Tracking (DABKT) model based on ability evaluation, which extends the hidden state by probabilistic modeling using KWA features on the basis of BKT so as to achieve comprehensive modeling of knowledge and ability, and introduces the difficulty feature to improve the predictive performance of the model.