Personalized Learning Recommendations Based on Feature Extraction and Attention Mechanisms

Abstract In recent years, the rise of online learning platforms has made personalized learning resource recommendation systems a key research focus. However, many existing algorithms fail to account for the evolving interests and needs of learners, limiting their effectiveness. To address this, we propose DynLearn-Adapt, a novel recommendation model that dynamically captures changes in student knowledge and behavior. First, the model uses convolutional neural networks to extract local learning features. It then employs a Transformer with a multi-head attention mechanism to model dependencies between skills, enhancing representational capacity. To track knowledge evolution and forgetting effects, DynLearn-Adapt integrates long- and short-term memory networks, enabling real-time updates to recommendations based on students? latest learning status. Additionally, positional encoding is introduced to better handle temporal information in learning sequences. Experiments conducted on two public datasets demonstrate that DynLearn-Adapt outperforms several baselines across multiple metrics, including accuracy, precision, recall, F1 score, and AUC, confirming its effectiveness and practical value.