An Agent-based Approach To Predict The Behaviors Of Learners With Difficulties Inside Human Learning Environments

Résumé: During their learning process, learners may encounter learning difficulties that may affect the quality of their academic outcomes. These difficulties could be triggered by external factors such as inadequate teaching content or internal factors closely related to some learners' specific characteristics. However, rather than looking for flaws in learners, it is more effective to explore extrinsic factors such as the form and relevance of teaching pedagogical content that is more adaptable to change and improvement than learner-specific factors. Struggling learners need different forms of support to learn more effectively and, if possible, catch up with their ordinary peers in terms of academic success. Nevertheless, it is essential to identify learners with difficulties at early stages to benefit from the appropriate support. However, before that, it is imperative to determine the signs and indicators to identify these learners in the face-to-face learning context in general and in e-learning environments in particular. This research is situated in this context and focuses on the learning difficulties faced by learners when using distance learning systems, as well as the intelligent tools available to help them overcome these difficulties. The use of distributed artificial intelligence techniques and, in particular, intelligent agents can solve the problem of detecting learners' learning difficulties and offer them the appropriate support at the right time. In recent years, new intelligent tools adopting new learning theories are continually being integrated into modern learning systems through predictive modeling used as Early Warning Systems (EWS), where one can identify and predict learners at risk in a given learning unit and inform both the teacher and the learners concerned. By collecting and analyzing learners' behavior through the traces left by them and using Distributed Artificial Intelligence (DMI) algorithms such as Multi-Agent Systems (MAS), it is possible to model, track, and monitor separately the current or even future behavior of each learner and identify which of them are doing well and which will face the likely difficulties providing valuable time to intervene and help these learners. Learners. To achieve these goals, a set of cognitive agents have been designed and implemented to detect learners' difficulties on the one hand and predict learners' failure or success on the other hand based on their behaviors. Prototypes validating the ideas proposed in this research work were developed and tested under real learning conditions. The results obtained are considered very promising and very encouraging.

Publié dans la revue: