New multi-agent approach for generating feedbacks base on Multiple Choice Questions

Abstract

Most intelligent tutorial systems promote the learning and resolution of exercises based on feedback in the form of advice, remarks, explanation…To always remain in self-assessment, our approach offers a new type of feedback in the form of multiple-choice questions applied in the field of algorithms (language c, java ...) dedicated for beginners in programming. The approach is based on the multi-agent model to have interaction between learners without the help of the teacher. There are three types of agents in our work, there is the learner agent who represents the learner, there is also the feedback agent who sends the MCQs to the learners who made mistakes and finally, the controller agent feeds the base of multiple-choice questions based on feedback from learners. The controlling agent compares the instructions of the learners with the instructions of the correct model based on the AST abstract syntax tree and detects errors which will be proposed as erroneous items (distractors) for the learners. We can say that this type of feedback is not direct like other work (advice, comments, explanation ...) but we always remain in the evaluation: MCQ, exercise ... As perspectives, we will focus mainly on the classification of LO learning objects in the form of an ontology to facilitate the use of data and the generation of multiple-choice questions at different levels. And also, we aim to develop a suitable platform which allows to define the agents and the messages to be exchanged between them to set up our system.