Sustainability Concerns in Open Source Software: A Multi-Domain Mining Study of GitHub Issues

Replication package for the paper "Sustainability Concerns in Open SourceSoftware: A Multi-Domain Mining Study of GitHub Issues", published at SBES 2026(40th Brazilian Symposium on Software Engineering). ## Overview This package contains the complete data and code for a Mining SoftwareRepositories (MSR) study of how sustainability concerns (energy efficiency,resource consumption, and performance) are reported and handled in open sourceissue trackers. The study mined 94,915 closed issues from 50 GitHub repositories spanning fivedomains (Systems, Backend, ML/Data Science, Frontend, and Mobile). A two-stepkeyword and noise filter reduced these to 6,849 candidates, which wereclassified using an LLM and validated through manual review, yielding 4,386sustainability-related issues. ## Contents Scripts (Python):- github_issue_miner.py: collects issues via the GitHub REST API and applies the keyword and noise filter- llm_classifier.py: classifies candidate issues- apply_corrections.py: applies manual corrections to the classifier output- analysis.py: produces the figures and supporting CSVs reported in the paper Data:- all_issues.zip: 94,915 closed issues (full corpus)- candidate_issues.zip: 6,849 issues retained after filtering- classified_issues.zip: 6,849 classified issues, 4,386 marked relevant- needs_manual_review.csv: 208 issues reviewed manually (Cohen's kappa = 0.746)- false_negative_sample.csv: 200 rejected issues reviewed for recall (false negative rate = 5.5%) Taxonomy:- keyword_patterns.csv: 10 categories and 105 regex patterns- classification_guidelines.md: decision rules used during manual review ## Reproducing the results cd data && unzip '*.zip' && cd .. python scripts/analysis.py Outputs are written to output/analysis/. Requires pandas, scipy, matplotlib,and seaborn. See README.md for the full pipeline, dataset relationships, anddata collection notes. ## Data collection notes Issues were collected on a fixed date; re-running the miner will produce alarger corpus, as GitHub issues accumulate over time. Issue histories aretruncated by two limits: an intentional cap of 2,000 issues per repository(reached by 39 of 50 repositories), and GitHub's pagination ceiling, whichstopped three repositories below that cap. Both are discussed as threats tovalidity in the paper. ## License CC BY 4.0 Outputs are written to output/analysis/. Requires pandas, scipy, matplotlib, and seaborn. See README.md for the full pipeline, dataset relationships, anddata collection notes. ## Data collection notes Issues were collected on a fixed date; re-running the miner will produce alarger corpus, as GitHub issues accumulate over time. Issue histories aretruncated by two limits: an intentional cap of 2,000 issues per repository(reached by 39 of 50 repositories), and GitHub's pagination ceiling, whichstopped three repositories below that cap. Both are discussed as threats tovalidity in the paper. ## License C