The Big Data era and intelligent educational systems have empowered personalized learning. As one of the most effective personalized learning tools, Recommender Systems (RS) are applied for student performance prediction, and personalized content replenishment for learning remediation. A wide variety of context-aware RS for personalized learning have been devised and implemented, adherent with student's learning contexts such as location, time, and activity. Due to the physical constraints, today's education is still carried out at schools, making classes the indispensable and easily achievable context. Leveraging such information can be beneficial for performance improvement and effective learning recommendation in common classroom settings. In this work, we propose a novel approach, 'Class Contextual Factor' (CCF)-based RS that synthesizes students' personal and class-level factors for better performances. More specifically, we first derive the CCF from a weighted Q-matrix to estimate students' mastery levels over KCs using an attribute-based recommendation technique. Then, we ensemble an item-based collaborative filtering algorithm for remedial exercise recommendation. By using a real world dataset from an online intelligent tutoring system, evaluations show that our CCF -based method outperforms the popular counterparts (i.e., IRT, RS with collaborative filtering), and is able to provide interpretable results for traceable learning remediation.