<p>E-learning platforms have increasingly incorporated recommender systems to deliver personalized learning experiences. But most existing systems are designed for general learners and often overlook accessibility needs and diverse learning preferences, particularly for Deaf and Hard of Hearing (DHH) learners. This limitation highlights the need for accessibility-aware personalization strategies in recommender systems. This study proposes a hybrid recommendation framework that integrates dynamic learner modelling to address the cold-start problem in DHH-focused e-learning environments, where interaction data is limited. The proposed method models learner characteristics, including modality preferences, accessibility needs, and communication preferences, along with structured representations of learning resources. By leveraging interaction data collected from DHH learners, the system adaptively updates learner profiles and generates content recommendations that align with both learner characteristics and interaction behaviour. Experimental evaluation shows that the proposed method improves the relevance and accessibility of recommended learning materials. The results indicate enhanced learner engagement, particularly in the early stages of interaction where user data is limited. Overall, this work contributes to the development of inclusive recommender systems by incorporating accessibility-aware learner modelling and dynamic adaptation, thereby supporting effective personalization for DHH learners.</p>