Amended feeding for Deep Knowledge Tracing
José Naranjo, Changsheng Zhu, Veronika Stoffová · 2023
Knowledge Tracing is the process of tracking and monitoring a learner’s knowledge of a particular subject over time to identify gaps in knowledge and adjust instruction accordingly. In this paper, we present a simple idea showing another insight into how to approach this problem to get different outcomes. This work is focused on the data that is fed into Deep Knowledge Tracing (DKT) model. We propose to add a pre-trained layer between the input and the hidden layer. The introduced layer is designed to produce a better representation of the questions that are fed to the model. That representation should conserve the correlation between the questions. That information provides a hint to the DKT model to enhance prediction accuracy. The experiments show that it is possible to extract the correlation between the exercises from the sequence of answers a learner gave to a random sample of the exercises.