Towards Improving Mathematics Learning Using the Deep Learning Models

Nowadays, the majority of intelligent tutoring systems (ITS) have significantly advanced in terms of capabilities and technologies. These systems have evolved to provide personalized and adaptive learning experiences for students. In mathematics, ITS can provide hints or messages to facilitate problem solving, but they do not provide feedback on educational concepts useful for resolving (such as theorems, definitions, etc.). Our methodology enables the generation of feedback in human learning environments. It makes it possible to predict the didactic concepts relevant to solving a given question Q(i) in an exercise based on existing or previously demonstrated data (Q(i-1), Q(i-2),…) using the Math-Bridge ontology. To extract knowledge from mathematical text, we use vectorization with SBERT models to calculate the semantic similarity between ontology concepts and exercises. We tested our approach on a set of exercises from mathematics teaching platforms and compared the number of concepts generated by our approach with those of experts. The results highlight the encouraging potential of our exercise solution method.