Using Possibilistic Networks to Compute Learning Course Indicators

AbstractE-learning systems generate more and more data that can be used to improve pedagogy. They can also provide a better understanding of a student’s learning style. As a result, it is possible to propose a differentiated pedagogy which takes into account learners’ needs. The aim of this research is to build course indicators by using expert knowledge in order to provide a synthesis of information about students. As knowledge is often imprecise and uncertain, we used possibility theory to represent knowledge through a possibilistic network. Firstly, we used a message passing algorithm to compute learning course indicators, then we proposed several improvements. Indeed, the use of uncertain gates allows us to generate automatically Conditional Possibility Tables (CPT) instead of eliciting all parameters. Next, we compiled the junction tree of the possibilistic network in order to improve computation time. We compared our compiling approach with message passing inference. A decision support system is generated automatically at the end of the computations. The indicators are presented in a decision support system in which color codes illustrate certainty.KeywordsCompiling knowledgeDecision MakingEducationPossibilistic networksPossibility theoryUncertainty