Here the authors examine if in Deep-IRT model they face the same problems as in the DKT model. These problems are: student performance decreases even if he/she makes a successful attempt and the estimated performances for different KCs are inconsistent over time. It was observed that prediction transitions were smooth while several irrelevant predictions were made.  

San Diego State University

1. Going Deeper in Difficulty Level
2. Going Deeper in Student Ability

Discussion (Deep-IRT: Make Deep Learning Based Knowledge Tracing Explainable Using Item Response Theory)

The authors used FSAIF1toF3 in order to evaluate estimated difficulty. They used this dataset as the publisher provided difficulty levels for each question. Another way to measure the difficulty of the question is to take into account how many students answered it correctly and how many didn't. The authors used only first attempt of a student to learn difficulty level and additionally they only considered question for which at least 10 students answered. The authors used PFA and entire student's learning trajectory to estimate difficulty level for the question. They only used questions that belonged to 5 subset of skills. As it was suggested it would interesting to find out if Deep IRT models give more accurate results than the traditional models. 

Learn Before

Related