Context-aware mobile learning is becoming important because of the dynamic and continually changing learning settings in learner's mobile environment, giving rise to many different learning contexts that are difficult to apprehend. To provide personalization of learning content, we aim to develop a recommender system based on semantic modeling of learning contents and learning context. This modeling is complemented by a behavioral part made up of rules and metaheuristics used to optimize the combination of pieces of learning contents according to learner's context. All these elements form a new approach to mobile learning.