Article Preview
TopIntroduction
The use of e-learning systems has increased in recent years due to the Covid-19 pandemic. As a result, deeper problems have been identified in these e-learning systems through their daily use despite their contribution in teaching as well as in industry (Chang, 2016). Thus, a superficial optimization of these connected environments is thus considered insufficient, as deep implications such as the psychological side of the learners combined in the case of engagement level learning must be considered.
There are several types of information that can provide the level of engagement of a student, such as indicators of course completion, number of interactions in the forum, etc. However, the latter do not reflect an essential aspect of the student, namely his emotional state, which must be deciphered in real-time just like in a traditional classroom face-to-face teaching. Fortunately, nowadays, serious technological advances can clearly predict the emotions experienced by a person, based solely on the facial analysis of this individual, using the photos provided by computer vision. Indeed, the latter allows collecting instantly a large number of photos of students during an online video conference. These photos provide a wealth of information about the student's condition and engagement.
Much research has already exploited this type of data to predict students' engagement (Chen et al., 2021; Liao et al., 2021; Delgado et al., 2021; Soffer & Cohen, 2019; Dhall et al., 2020) using very powerful Machine Learning (ML) algorithms. However, in most of the research, the prediction of engagement is not accompanied by propositions that prompt the teacher or learner to take actions according to the considered situation favoring the boost of engagement.
Although there are other works on the optimization of learning systems that consider emotional aspects such as satisfaction, confirmation, enjoyment (Aulia Winarno et al., 2020), or even the students' conditions (Elshaer & Sobaih 2022). Nevertheless, the aspect of formulating suggestions related to the emotional state of the students remains poorly explored in the literature.
Similarly, there is a lot of research work on recommendation systems in e-learning. Indeed, these cover several different themes such as personalized content delivery (Chang et al., 2022; Gomede et al., 2021), or personalized learning activities (Bhaskaran et al., 2021). Nevertheless, little thought has been given to the scalability of learning systems.
To this end, the authors of this paper propose the development of an Intelligent Recommendation System for Optimizing the quality of online Learning and Teaching environments (IRSOLT), which considers students’ engagement. This system will be able to predict students' engagement states and produce concise suggestions adapted to each situation. The same system produces a new learning indicator at the end of a videoconference course that will be stored in the learning system database. This indicator will later allow determining the state of engagement of the learners and improving it either in real-time or even deferred, thus allowing the scalability of this proposed system.
For this purpose, this paper is organized into several sections. In the second section, a literature review will be exposed. The third section will be dedicated to the working methodology and the material used. The IRSOLT rules database will be presented in the fourth section. The fifth section will be dedicated to the evaluation metrics of the proposed system. The results obtained and their discussion will be presented in sections six and seven. Section eight will be devoted to the limits of the proposed system. Finally, the authors of this paper will sum up their research paper with a conclusion and perspectives.