<p>Recent advances in deep learning have significantly enhanced the personalization of educational resources. However, traditional recommendation systems often struggle with critical limitations such as the cold-start problem, over-specialization, and an inability to capture learners' implicit cognitive needs. To address these issues, we propose a personalized recommendation system that leverages the TabTransformer neural model to predict learning styles. Our approach uniquely integrates diverse data, combining structured numerical metrics with unstructured textual data from learner comments. By applying Natural Language Processing (NLP) techniques, we extract rich, implicit signals about learner preferences and difficulties that are often missed by numerical data alone. The primary contributions of this work are the accurate identification of learners' cognitive preferences using behavioral data and the dynamic adaptation of recommendations based on the Felder-Silverman Learning Style Model (FSLSM). This multimodal framework advances pedagogical personalization and supports more informed educational strategies.</p>