Abstract:

Traditional learning systems ("all to all" system) provide the same pedagogical content to all learners, regardless of their profile, prior knowledge or preference. Such an approach proves to be ineffective especially when dealing with learners of different backgrounds and needs. Today, with the new adaptive learning approach (e-learning; m-learning), it is now possible to customize educational content based on individual needs of each learner. In relation to this new approach, several research studies have been carried out precisely on: Programmed teaching (Skinner and de Crowder 1954), which is based on the division of the material to be taught by granules, and Computer Assisted Teaching, which allows the acquisition of concepts at the learners' own pace. Currently, this new vision is oriented towards the design and development of new systems commonly known as "Computing Environment for Human Learning", of which Dynamic Adaptive Hypermedia Systems appear to be the most widely used ones. Their main objective is to script learner's pedagogical activity and thus monitor and manage their learning process (Mellet, 2006; Settouti, 2006). In this regard, the literature review allows us to identify a set of systems, namely: SMEXWEB system (Koch, 2000); KOD system (Sampson, 2002); MEDYNA system (Behaz, 2008); OrPAF system (Yessad, 2009), etc. However, an extreme number of new adaptive teaching and learning approaches based on advanced technologies addressing parallel intelligent education - an artificial education system (Ying Tang et al. 2020).

In order to remedy certain shortcomings of these systems, in particular the division of the content based on the detriment of its quality, the absence of tools to guide learners and to improve the feasibility of these systems, we recommend the optimal technique (Q-function) based on learning by reinforcement of Artificial Intelligence through a multi-agent architecture deployed on the JADE Framework in one hand, and on automatic learning using the K_Means clustering algorithm on the other hand. Our approach aims at extending these standards by integrating a coherent behavioral model based on the AUML formalism (UML Agent). In other words, the expected system must be capable of dynamically adapting a learning process to any given learner on the basis of the assembly existing resources. Thus, we model this method by a graph representing learning process, in which vertices are objects learning (OL) (Courses, Videos, etc.) and in which arcs (evaluated by rewards/sanctions) are actions (circuits taken by the learners) to move from one object learning to the next one . This tree diagram not only will allow us to better adapt and personalize pedagogical content and to categorize learners according to their achievements, but also to consider the learning processes to be followed by other learners with the same profile and prerequisites.

Keywords:

Adaptive Learning, Reinforcement Learning, Clustering Algorithm, Q-Function, Objects Learning, Automatic Learning.