Reproduce “Assessing Consensus of Developers' Views on Code Readability” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Agnia Sergeyuk, Olga Lvova, Sergey Titov, Anastasiia Serova, Farid Bagirov, Timofey Bryksin
Researchers from JetBrains Research demonstrated that Java developers with similar professional backgrounds show statistically significant consensus in their assessments of code readability and its related aspects. The study identified specific code characteristics, such as "Code Length" and "Understandable Goal," that are highly correlated with overall readability and were consistently rated, providing concrete learning objectives for fine-tuning Large Language Models to generate more human-readable code.
Key insights
- Developer consensus on code readability is achievable and statistically significant within a group sharing similar professional backgrounds and potentially a common codebase.
- Specific code characteristics previously identified through the Repertory Grid Technique are indeed representative of developers' notions of overall code readability.
- The findings provide a foundation for defining concrete "learning objectives" for LLMs, enabling them to generate code that is not only correct but also aligned with human readability standards.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.