Problem Solving-Oriented Programming Knowledge Tracing from Behavior to Thought

Abstract

Programming knowledge tracing (programming KT) aims to analyze the dynamic programming states in solving problems based on historical behaviors and predict future performance. In programming, a student’s thought process can lead to multiple solutions for the same problem. However, current programming KT models attribute learners’ responses only to knowledge mastery and ability, overlooking thought factors, which creates a contradiction between data fitting and the rationalization of the model inference process. To address this, we propose a problem solving-oriented programming KT (SPKT) method that incorporates programming knowledge, computational thinking, and solving ability to improve attribution accuracy. Specifically, we designed a dual-channel attention network based on the principle of initiative gain to retrieve knowledge. Additionally, we utilized the edit tree distance algorithm to capture fine-grained trajectory representations and employed a redress mechanism with gating to update abilities based on code information. Experiments demonstrate SPKT’s superiority.