Skip to main content
@shmVirus

publications

ExplainCode: Towards Explainable Automated Feedback for Introductory Programming

Sabbir Hosen Mamun†, Priya S. Nair, Marcus E. Hoffmann, Beatriz M. Cardoso† Corresponding author

ACM/IEEE International Conference on Software Engineering, pp. 1432–1444, Lisbon, Portugal · ACM

✦ ACM SIGSOFT Distinguished Paper Award

Conference202421 citations
PDFDOICodeVideoCite

Abstract

Automated feedback tools for introductory programming typically report what is wrong but not why, limiting their educational value. ExplainCode combines lightweight program analysis with a retrieval-augmented language model to generate student-facing explanations grounded in common misconception taxonomies. In a randomised controlled study (n = 96), learners who received explanatory feedback achieved significantly higher scores on the subsequent assignment (M = 72.4 vs. 61.3, t(94) = 3.12, p < 0.01) and rated the feedback as more actionable (4.2 vs. 2.8 on a 5-point scale). Qualitative coding of student responses reveals that explanation quality — not mere correctness — is the strongest predictor of follow-through.

Cite this work

@inproceedings{Mamun2024ExplainCode,
  title = {ExplainCode: Towards Explainable Automated Feedback for Introductory Programming},
  author = {Sabbir Hosen Mamun and Priya S. Nair and Marcus E. Hoffmann and Beatriz M. Cardoso},
  year = {2024},
  booktitle = {ACM/IEEE International Conference on Software Engineering},
  pages = {1432–1444},
  address = {Lisbon, Portugal},
  publisher = {ACM},
  series = {ICSE '24},
  doi = {10.1145/3597503.3608128},
  abstract = {Automated feedback tools for introductory programming typically report what is wrong but not why, limiting their educational value. ExplainCode combines lightweight program analysis with a retrieval-augmented language model to generate student-facing explanations grounded in common misconception taxonomies. In a randomised controlled study (n = 96), learners who received explanatory feedback achieved significantly higher scores on the subsequent assignment (M = 72.4 vs. 61.3, t(94) = 3.12, p < 0.01) and rated the feedback as more actionable (4.2 vs. 2.8 on a 5-point scale). Qualitative coding of student responses reveals that explanation quality — not mere correctness — is the strongest predictor of follow-through.},
}