The dataset consists of human-annotated code snippets sampled from real open-source projects and labeled for perceived readability. It is the largest such dataset collected for software readability research to date.

📦 Dataset Summary

Property	Value
Total snippets	360
Programming languages	Java, Python, CUDA
Snippet lengths	~10, ~30, ~50 lines
Human annotations	≈ 5,468 annotators
Total ratings	≈ 76,741 readability votes
Rating scale	1–5 Likert (1 = very unreadable, 5 = very readable)
Annotator background	students + industry (1,000+ with 5+ years professional experience)
Source projects	30 open-source repositories

The dataset was collected through an IRB-approved online survey. Each participant viewed 20 random snippets and rated their readability. Samples were drawn automatically and uniformly from repositories, without manual curation, to avoid bias.

🧠 Tasks Supported

This dataset supports several research directions:

🎯 Core Tasks

readability prediction (regression or classification)
feature engineering for code comprehension
cross-language readability comparison

🌍 Languages & Projects

Languages sampled:

Java (large, object-oriented)
Python (indentation-sensitive)
CUDA (GPU programming)

Each sourced from 10 real open-source repos (30 total), including widely-used projects like:

Liferay Portal
SQuirreL SQL Client
Docutils
GPUMLib

(see paper for full list)

📝 Annotation Protocol

ratings used a 1–5 Likert scale:
- 1 → very unreadable
- 5 → very readable
code was syntax-highlighted in survey
annotators could revise earlier answers
snippets were shown without needing to be syntactically complete
annotators provided experience metadata

🤝 Citation

If you use this dataset, please cite:

@inproceedings{dorn2012readability,
  title={A general software readability model},
  author={Dorn, Jonathan and Weimer, Westley},
  booktitle={International Conference on Software Engineering},
  year={2012}
}

🙌 Acknowledgements

Special thanks to the thousands of anonymous survey respondents, Udacity participants, and reddit programmers who contributed ratings.

Contact

If you have questions or would like to extend the dataset, feel free to open an issue or discussion in the repo.

Downloads last month: 19