Curating GitHub for engineered software projects
Bibliographic record
Abstract
Software forges like GitHub host millions of repositories. Software engineering researchers have been able to take advantage of such a large corpora of potential study subjects with the help of tools like GHTorrent and Boa. However, the simplicity in querying comes with a caveat: there are limited means of separating the signal (e.g. repositories containing engineered software projects) from the noise (e.g. repositories containing home work assignments). The proportion of noise in a random sample of repositories could skew the study and may lead to researchers reaching unrealistic, potentially inaccurate, conclusions. We argue that it is imperative to have the ability to sieve out the noise in such large repository forges. We propose a framework, and present a reference implementation of the framework as a tool called reaper, to enable researchers to select GitHub repositories that contain evidence of an engineered software project. We identify software engineering practices (called dimensions) and propose means for validating their existence in a GitHub repository. We used reaper to measure the dimensions of 1,994,977 GitHub repositories. We then used the data set train classifiers capable of predicting if a given GitHub repository contains an engineered software project. The performance of the classifiers was evaluated using a set of 200 repositories with known ground truth classification. We also compared the performance of the classifiers to other approaches to classification (e.g. number of GitHub Stargazers) and found our classifiers to outperform existing approaches. We found stargazers-based classifier to exhibit high precision (96%) but an inversely proportional recall (27%). On the other hand, our best classifier exhibited a high precision (82%) and a high recall (83%). The stargazer-based criteria offers precision but fails to recall a significant potion of the population.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.008 | 0.054 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.024 | 0.016 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.002 | 0.006 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".