Adaptive learning approach aims to automatically generate learning paths that meet the specific needs of each learner according to their individual differences in learning preferences, goals, abilities, knowledge backgrounds, etc. In this paper, we propose a deep knowledge tracking model FDKT-ED that integrates the difficulty of exercises and forgetting behavior, and construct an adaptive dynamic learning path recommendation model RL4ALPR based on reinforcement learning based on FDKT-ED to realize dynamic learning path planning and goal motivation. The AUC and F1-Score of FDKT-ED model on 2 experimental datasets reach the average values of 71.03% and 75.68%, both of which are better than other comparative models, indicating that the model can realize in-depth knowledge tracking of students and improve the accuracy of the task of predicting students’ future answer performance. Meanwhile, the proposed RL4ALPR model gives test score improvement of about 0.35 and 0.41 from the environment models KSE and KEE, respectively, and its recommended results are most consistent with the learning logic as perceived by human experts, which is better than the other comparative models in terms of cognitive state improvement and learning logic. In addition, the results of the teaching experiment show that the experimental class using the learning path planning and goal motivation strategy of this paper outperforms the control class using the traditional teaching strategy in terms of reading, translation, writing and their total scores in the performance of university public English, which corroborates the feasibility of the modeling strategy of this paper in enhancing the effectiveness of university public English teaching.