In online education platforms, personalized recommendation systems are crucial to improving user experience. This study used the long short-term memory (LSTM) network in deep learning technology to build a model that can provide personalized learning resource recommendations by analyzing user behavior data. The model not only includes an embedding layer to capture user characteristics but also deeply mines the long-term dependencies of user behavior through multiple LSTM layers to achieve more accurate recommendation results. Deep learning containing embedding layers and multiple LSTM layers was studied. The experimental results showed that the LSTM model can effectively improve the user experience of online learning platforms, with the highest user satisfaction reaching 96.9 points. Moreover, personalized recommendation algorithms based on LSTM have advantages in click-through rate, recommendation relevance, coverage, and recommendation time. The recommendation system using the LSTM model significantly improves user satisfaction and learning motivation.