Personalized learning paths can take into account the individual differences of different learners’ abilities, knowledge backgrounds and knowledge mastery levels based on big data and provide targeted learning sequences for different learners to achieve the purpose of improving teaching efficiency. In this paper, we construct a personalized learning path optimization model that includes a neural network collaborative filtering layer, a cognitive level diagnosis layer, and a learning path generation layer. The nonlinear relationship between learners and knowledge points is represented by obtaining interaction information between learners and knowledge points, so that the model can better extract the vector features of learners and knowledge points. Teaching practice shows that the experimental class under this learning path improves by 25.82 points, which is significantly larger than the control class, and students in all segments make significant progress. There are no students with more than 80 points in the pre-test, while there are as many as 11 students with more than 90 points at the end of the semester, which verifies the effectiveness of the model in improving the efficiency of teaching and meeting the needs of personalized learning. This study provides useful insights for making scientific teaching decisions, objectively evaluating student learning outcomes, and improving teaching efficiency.