<p>Traditional recommendation systems often reach their limits when confronted with the diversity of learner profiles. To overcome this challenge, we propose a personalized and dynamically adaptive document ranking approach. This study introduces a two-module architecture designed to develop an adaptive and personalized recommendation system that supports learner progression through content tailored to their evolving profiles. The first module, previously validated in earlier work, enhances retrieval performance by optimizing indexing and integrating semantic features. It forms the foundation of the system by ensuring the relevance and significance of returned documents. The present work focuses on the second module, which aims to dynamically adapt document ranking throughout the learning session according to learners’ expertise levels with respect to the queried topic. Implicit learning traces are analyzed to characterize each learner based on their interests, preferences, expertise level, and role (novice or expert) in the search process. Semantic and contextual aspects are also incorporated to refine the result ranking based on user profiles. To achieve this adaptation, we design a deep learning architecture based on a dual recurrent encoder network, following a vector embedding phase for both queries and documents. This modeling approach captures the query context, semantic content, and learner profile simultaneously. The proposed system was evaluated on the <i>Jamalon Arabic Books</i> dataset using simulated session logs.</p>