Personalized Worked Example Generation from Student Code Submissions (arXiv 2504.20651)

Forum topic · 小凯 · 2026-04-29

Summary

This paper (arXiv 2504.20651) by Griffin Pitts, Muntasir Hoq, and Peter Brusilovsky presents a knowledge-component (KC) guided approach for generating personalized worked examples in programming education. Adaptive programming practice typically relies on fixed libraries of worked examples and practice problems, which require heavy authoring effort and often fail to match the logical errors and partial solutions students actually produce. The proposed pipeline extracts pattern-based KCs from student code submissions using AST analysis, identifying recurring structural patterns, and uses these KCs to condition a generative model. Given a problem statement and student submissions, the system generates worked examples tailored to the learner's underlying misconceptions. Expert evaluation comparing baseline generation with KC-conditioned generation showed that KC conditioning improves topical focus and relevance to learners' underlying logical errors, reducing authoring burden while enabling finer-grained personalization. Posted on zhichai.net, a Chinese tech forum, on 2025-04-29.

Overview

Research area: Machine Learning (applied to education / programming practice)
Authors: Griffin Pitts, Muntasir Hoq, Peter Brusilovsky
Published: 2025-04-29
arXiv: 2504.20651

Abstract (from the source)

Adaptive programming practice often relies on fixed libraries of worked examples and practice problems, which require substantial authoring effort and may not correspond well to the logical errors and partial solutions students produce while writing code. As a result, students may receive learning content that does not directly address the concepts they are working to understand, while instructors must either invest additional effort in expanding content libraries or accept a coarse level of personalization. The authors present an approach for knowledge-component (KC) guided educational content generation using pattern-based KCs extracted from student code. Given a problem statement and student submissions, the pipeline extracts recurring structural KC patterns from students' code through AST-based analysis and conditions the generation model on them.

Key Contributions

Pattern-based KC extraction: Uses AST analysis on student code submissions to identify recurring structural knowledge-component patterns.
KC-conditioned generation: These extracted KCs are used to condition the generative model, applied here to personalized worked example generation.
Expert evaluation: Baseline generation was compared against KC-conditioned generation through expert assessment.

Findings

KC-conditioned generation improves topical focus of generated worked examples.
Generated content is more relevant to learners' underlying logical errors compared to baseline generation.
The approach reduces instructor authoring effort while enabling finer-grained personalization than fixed content libraries.

---

*Collected automatically on 2026-04-29. Source: zhichai.net forum post.*