PAEGM: Enhancing Student Performance Prediction in Programming Learning Systems

Traditional programming learning systems face issues such as inaccurate knowledge state estimation, insufficient sequence processing capability, and difficulties in modeling local dependencies, with existing methods often overlooking the complex relationship between student responses and code structure, resulting in predictions lacking personalization. To address these challenges, this paper proposes enhancing Student Performance Prediction in Programming Learning Systems(PAEGM), which effectively improves prediction accuracy and model interpretability by incorporating both positive and negative knowledge relationships as well as the temporal evolution of knowledge states. Specifically, the PAEGM model enhances the knowledge state estimation process by introducing the Polarity-Aware Linear Attention mechanism (PolaLinearAttention) and a dual representation module. The dual representation module combines natural language processing with code structure analysis, improving the model's ability to understand student inputs and generating personalized predictions based on the student's evolving knowledge. Experimental results demon strate that PAEGM significantly improves the accuracy of student performance prediction, showcasing its potential for application in intelligent tutoring systems and online learning platforms.