LearnClusterVis: A Framework for Clustering-driven Visual Analysis of Programming Learners' Learning Process

The rapid growth of online grading systems (commonly referred to as online judge systems in programming education) provides valuable opportunities to analyze programming learners' processes, but the complexity of such datasets poses significant challenges for instructors lacking specialized analytical techniques. Furthermore, it remains a significant challenge for instructors to effectively identify priority learner groups that require targeted attention and to make informed educational decisions within classroom contexts. To address these challenges, we introduce LearnClusterVis, a clustering-driven visual analysis framework designed to uncover behavioral patterns and developmental trajectories in programming learners' activities. LearnClusterVis is highly extensible and can be applied to various online grading systems. LearnClusterVis leverages learners' submission records to generate customizable visual analysis interfaces, enabling instructors to explore learning patterns, identify learner clusters, monitor progress, deliver personalized interventions, and evaluate the rationality of questions across knowledge domains. The case studies, which implemented the framework using data from two distinct online grading systems, demonstrate its effectiveness and scalability.