Personalized learning, in which learners set their own pace and select their own resources according to their own learning needs and characteristics, is the trend of Chinese education and teaching. In this paper, we design a personalized teaching path recommendation model for Chinese education based on reinforcement learning. The knowledge tracking prediction model LTKT is designed to integrate multiple knowledge points as information dimensions for model learning in the data preprocessing stage. The sparse self-attention mechanism is introduced into the encoder and decoder structure and embedded with location coding containing absolute and relative distances to enhance the model’s perception of location information. Finally, the RL4ALPR algorithm is designed to model the changing knowledge level, the candidate learning item filtering algorithm is used to narrow down the scope of the recommended learning items, the reinforcement learning algorithm assumes the role of a recommender, and the degree of change in the knowledge level of the learner is regarded as a reward for the improvement of the reinforcement learning recommendation strategy. Simulation experiments are conducted on datasets such as ASSISTments and compared with baseline models such as KNN, GRU4Rec, Random, etc. The model in this paper has an F1 value and an AUC of 0.635 and 0.956 respectively in the evaluation of learning effect, which are the highest among the models. The study makes a useful exploration for the informatization of Chinese education and teaching.