The advantage of face-to-face teaching is the teacher's ability to naturally assess the engagement of his students, while adapting his course to their training needs. Whereas, in an online teaching environment, the identification of the learners' engagement becomes an arduous task involving considerable effort on behalf of the teacher negatively impacting the quality of his rendering. Therefore, the authors of this paper focus on the development of a new intelligent recommendation system for the optimization of the quality of online learning and teaching. This system provides suggestions to teachers or students based on their state of engagement. Moreover, the system developed is based on a deep learning model, the convolutional neural network, which has confirmed its reliability with an accuracy of 88%. Finally, the future scalability of the recommendation system is ensured through the introduction of a new learning indicator that can be used for further predictions.