Deep knowledge tracing for free-form student code progression

Abstract

Knowledge Tracing, and its recent deep learning variants, have made substantial progress in modeling student knowledge acquisition through interactions with coursework. In this paper, we present a modification to Deep Knowledge Tracing to model student progress on coding assignments in large-scale computer science courses. The model takes advantage of the computer science education context by encoding students’ iterative attempts on the same problem and allowing free-form code input. We implement a workflow for collecting data from Jupyter Notebooks and suggest future research possibilities for real-time intervention.