Long short-term attentional neuro-cognitive diagnostic model for skill growth assessment in intelligent tutoring systems

Measuring student growth and providing diagnostic feedback are core components of cognitive diagnostic assessment. However, most current cognitive diagnostic models solely rely on data from a single occasion to diagnose student skill states, overlooking the substantial long-term information encapsulated in the learning history from multiple occasions. In this paper, we propose a long short-term attentional cognitive diagnostic (LS-ENCD) model for skill growth assessment in intelligent tutoring systems. Specifically, we first embed exercise and student features into high-dimensional vectors. Then, we use a measurement module with a bilayer architecture to establish the interaction between students and exercises, considering guessing and slipping factors. To capture long short-term dependencies on historical data, we design the long short-term learning transfer module based on the attention mechanism, which computes state transfer weights by incorporating occasion time and mastery state. Finally, extensive experimental results on four public datasets demonstrate the superiority and good interpretability of our proposed model.