Using Deep Knowledge Tracing to Predict a Simulated Distance Measure

Abstract

Knowledge tracing concerns keeping track of the skills a student has learned over time by doing exercises that are labelled with a skill. Various ways of knowl- edge tracing include Bayesian Knowledge Tracing (BKT), Performance Factor Analysis (PFA) and Deep Knowledge Tracing (DKT). DKT uses a Recurrent Neural Network (RNN) in combination with Long Short Term Memory units to predict the performance of students. The binary output of this RNN depicts the performance of a student. If a student is wrong, then this output will not give information about how wrong a student was. Therefore, a distance measure will be added as a second output to the RNN. This distance measure is de ned as the distance between the real answer and the one given by the student. In this thesis a DKT model is build with this simulated distance measure as addi- tional input and output. Unfortunately, the accuracy of these predictions was insu cient. This is probably due to a wrong interpretation of the nature of this problem and therefore future research is needed.