Evaluating Static and Process Evidence for Code Authorship in Programming Education

2026-07-09 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, extended

Summary

A study by Marek Horváth (2026) evaluates source-code authorship in programming education, contrasting it with programming-contest datasets. It reveals that contest data, like *Kick Start*, yield high attribution performance with a mean top-1 of 0.938. In contrast, educational datasets show substantially lower initial attribution performance, with a mean top-1 of 0.094. Crucially, incorporating repository-visible process features significantly improves educational attribution, raising the mean top-1 to 0.233 and mean pairwise verification ROC–AUC from 0.556 to 0.752. The research emphasizes that the measured signal strength depends heavily on the production context, and process patterns effectively complement the weaker final-code signal found in educational settings, positioning these models as instructor-mediated decision support rather than independent proof of authorship.

Key takeaway

For programming instructors evaluating student submissions, recognize that code authorship models are decision support tools, not definitive proof. If code similarity is inconclusive, integrate repository-derived process features like commit history and work patterns into your assessment. This approach can improve the ranking of plausible authors and help prioritize cases for human review, offering a more nuanced assessment than static code analysis alone.

Key insights

Process features significantly enhance code authorship attribution in educational settings where static code signal is weak.

Principles

• Authorship signal is context-dependent.
• Educational programming style is less stable.
• Process data complements static code analysis.

Method

Combine static code features with repository-derived process features (commits, timing, change volume) using random forest for attribution and logistic regression for verification.

In practice

• Integrate commit history into plagiarism checks.
• Prioritize student submissions for review.
• Analyze work patterns for feedback.

Topics

• Code Authorship
• Programming Education
• Stylometry
• Process Features
• Static Code Analysis
• Academic Integrity

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• HybridCodeAuthorship: A Benchmark Dataset for Line-Level Code Authorship Detection — Industry codebases require fine-grained, line-level detection of AI-generated code for risk and productivity analysis.
• Dream at SemEval-2026 Task 13: SALSA for Single-Pass Machine-Generated Code Detection — SALSA enables robust, single-pass detection of machine-generated code, achieving strong out-of-distribution generalization.
• MALTO at SemEval-2026 Task 13: Detecting Human, AI, and Hybrid Code via Hard Negative Mining and Curriculum-Driven Ensembles — An ensemble of specialized models with advanced training techniques robustly detects human, AI, and hybrid code origins.
• Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code — Citation annotations in LLM-generated code trade output determinism for critical verifiability and automated hallucination detection.
• ‘Obvious markers of AI’: doubts raised over winner of short story prize — The definitive detection of AI-generated creative work remains elusive, posing challenges for literary awards and publishers.
• Project-wise Comparison of Software Birthmarks Using Weighted Partial Similarity — Project-wise software birthmark comparison requires weighting module similarities and focusing on top matches to detect partial reuse and mitigate incidental similarity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.