Personalized Worked Example Generation from Student Code Using Pattern-Based KCs
A research paper on arXiv (2604.24758v2) introduces a method for generating personalized worked examples from student code submissions. The approach uses pattern-based knowledge components (KCs) extracted via AST-based analysis to condition a generative model. This addresses the limitations of fixed example libraries, which require extensive authoring and often fail to match students' logical errors and partial solutions. The pipeline takes a problem statement and student submissions, extracts recurring structural KC patterns, and generates tailored worked examples. The study compares this method against baselines, aiming to improve adaptive programming practice by directly targeting concepts students are working to understand.
Key facts
- arXiv paper 2604.24758v2
- Approach uses pattern-based knowledge components (KCs)
- KCs extracted from student code via AST-based analysis
- Generative model conditioned on KCs
- Addresses limitations of fixed example libraries
- Pipeline takes problem statement and student submissions
- Generates personalized worked examples
- Study compares against baselines
Entities
Institutions
- arXiv