Developer experience · Machine-learning recommender (RecGFI) using content, background, and dynamics features; historical-state reconstruction and maintainer validation.

Recommending good first issues in GitHub OSS projects

Peer-reviewed research · 2022

Ana Paiva, Rui Abreu, Abhik Roychoudhury, Margaret-Anne Storey

Original paper ↗

Summary

RecGFI reached up to 0.853 AUC, outperformed alternatives, and among reported recommendations 16 were confirmed real GFIs, with five resolved by newcomers.

For your marketing

Possible founder application

Automate candidate-task discovery, then have maintainers validate clarity, scope, and dependencies before labeling.

Reported result

What the study found

RecGFI reached up to 0.853 AUC, outperformed alternatives, and among reported recommendations 16 were confirmed real GFIs, with five resolved by newcomers.

Research question

What the authors investigated

Can a model identify suitable Good First Issues and reduce maintainer burden?

Design

Machine-learning recommender (RecGFI) using content, background, and dynamics features; historical-state reconstruction and maintainer validation.

Sample and setting

53,510 resolved issues from 100 GitHub projects; latest recommendations were reported to maintainers.

Scope and transfer

Limitations to keep in view

Performance is based on 100 projects and historical labels; the paper does not establish long-term retention from recommendations.

Research sources

Publisher/authoritative landing page and abstract/full text fetched: https://dl.acm.org/doi/10.1145/3510003.3510196 Firecrawl research job 01a14366-d263-71d0-bc44-b95482987b7b returned this citation and metadata.

Study strength

3.5/5 Abstract-level evidence is useful for framing questions; verify full methods before transfer.

How this was assessed

The 1–5 evidence-check score is a practical reading aid, not a formal peer review or a probability that a conclusion is true. It combines design strength (2 points), sample reporting (1), comparison/control clarity (1), and how much of the paper could be checked (1). Access affects how confidently this directory can assess a study, not the study's underlying quality. Blinding is reported separately because it is not relevant to every design.

Design transparency · 2/2

Based on the reported study design in the retrieved source.

Sample and setting · 1/1

The sample context is stated; a numerical count is preserved only when reported.

Outcome clarity · 1/1

The finding is summarized from publisher or article metadata.

Access for checking · 0.5/1

Only abstract or bibliographic access was confirmed.

Double-blind? The retrieved publisher record does not establish whether participants or outcome assessors were blinded.

Read the paper