Context-aware knowledge tracing integrated with the exercise representation and association in mathematics
Abstract: Influenced by Covid-19, online learning has become one of the most important forms of education in the world. In the era of intelligent education, knowledge tracing(KT), which can estimate and predict a student’s level of knowledge mastery, can provides excellent technical support for individualized teaching. In our proposed method, we come up with a new knowledge tracing method that integrates mathematical exercise representation and association of exercises(ERAKT). In the aspect of exercise representation, we represent the multi-dimensional features of the exercises, such as formula, text and associated concept, by using ontology replacement method, language model and embedding technology, so we can obtain the unified internal representation of the exercises. Besides, we utilize the bidirectional long short memory neural network to acquire the item association between exercises, so as to predict his performance in future exercise. Extensive experiments on a real dataset, Eanalyst-math, clearly proved the effectiveness of ERAKT method. Experiments also verified that adding multi-dimensional features and exercise association can indeed improve the accuracy of prediction.