With the rapid development of digital education, college English teaching has put forward higher requirements for the accurate design and dynamic adaptation of learning paths. Focusing on the path recommendation task driven by multi-source learning data, this paper constructs a technical framework covering data collection and preprocessing, learner portrait modeling, deep feature extraction, path generation, resource matching, dynamic adjustment and teaching feedback closed loop. Intelligent recommendation and continuous update of learning paths are realized. The experimental results show that the Precision, Recall and F1-score of the proposed method reach 0.846, 0.821 and 0.833, the Top-10 hit rate reaches 0.862, the path generation accuracy and knowledge coverage rate reach 0.852 and 0.874, respectively. After feedback-driven adjustment, the task accuracy is increased to 84.9%, and the path deviation index is reduced to 0.121. The research shows that this method can improve the accuracy, coherence and dynamic adaptation ability of college English learning path, and provide technical support for the intelligent transformation and accurate implementation of college English teaching.
