This study proposes a personalized learning path recommendation model that integrates deep learning, fuzzy cognitive diagnosis, and Bayesian methods to achieve large-scale personalized teaching. Firstly, construct a multidimensional learner profile and extract features such as attitude, engagement, and focus from behavioral data; Using fuzzy cognitive diagnosis (FuzzyCDF) to quantify students’ knowledge ability and mastery level of knowledge points, distinguishing the correlation and compensation effects between subjective and objective test questions; Secondly, a knowledge point relationship network is constructed based on Bayesian networks, combined with ant colony algorithm to generate multi-objective optimization learning paths. Finally, the index grading was completed using the natural breakpoint method (JNB), and a simulation experiment was conducted with 125 architecture students. Three sets of quasi experiments were designed with 165 advanced mathematics students to verify the effectiveness. The results showed that the recommendation path recognition degree improved by about 20% with the integration of cognitive diagnosis, the optimization of learning time exceeded 35%, and the learning enthusiasm increased by 10% to 17%; The average post test score of experimental group B reached 87.98 points, significantly higher than the traditional recommendation group (76.47 points) and the control group (69.56 points), and the score differentiation was significantly reduced, providing a feasible solution for the intelligent education system.
