Skip to main content
ConductScienceScienceDex
zenodoopen

88.6 Million Developer Comments from GitHub

<p><em><strong>Description</strong></em></p> <p>This is a collection of developer comments from GitHub issues, commits, and pull requests. We collected 88,640,237&nbsp;developer comments from 17,378 repositories. In total, this dataset includes:</p> <ul> <li>54,252,380 issue comments (from 13,458,208 issues)</li> <li>979,642 commit comments (from&nbsp;49,710,108 commits)</li> <li>33,408,215 pull request comments (from&nbsp;12,680,373 pull requests)</li> </ul> <p><strong>Warning:</strong> The uploaded dataset is compressed from <strong>185GB</strong>&nbsp;down to 25.1GB.</p> <p>&nbsp;</p> <p><em><strong>Purpose</strong></em></p> <p>The purpose of this dataset (corpus) is to provide a large dataset of software developer comments (natural language) for research. We intend to use this data in our own research, but we hope it will be helpful for other researchers.</p> <p>&nbsp;</p> <p><em><strong>Collection Process</strong></em></p> <p>Full implementation details can be found in the following publication:</p> <ul> <li>Benjamin S. Meyers.&nbsp;<a href="https://scholarworks.rit.edu/theses/11609/">Human Error Assessment in Software Engineering.</a> Rochester Institute of Technology. 2023.</li> </ul> <p>Data was downloaded using&nbsp;<a href="https://docs.github.com/en/graphql">GitHub's GraphQL API</a> via requests made with Python's <a href="https://docs.python-requests.org/en/master/">requests</a> library. We targeted 17,491 repositories with the following criteria:</p> <ul> <li>At least 850 stars.</li> <li>Primary language in the Top 50 from the <a href="https://www.tiobe.com/tiobe-index/">TIOBE Index</a> and/or listed as "popular" in GitHub's advanced search. Note that we collected the list of languages on August 31, 2021.</li> </ul> <p>Due to design decisions made by GitHub, we could only get a list of at most 1,000 repositories for each target language. Comments from 113 repositories could not be downloaded for various reasons (failing API queries, <a href="https://docs.python.org/3/library/json.html#json.JSONDecodeError">JSONDecoderErrors</a>, etc.). Eight target languages had no repositories matching the above criteria.</p> <p>After collection using the GraphQL API, data was written to CSV using Python's <a href="https://docs.python.org/3/library/csv.html?highlight=csv%20writer#csv.writer">csv.writer</a> class. We&nbsp;<strong>highly recommend</strong>&nbsp;using Python's <a href="https://docs.python.org/3/library/csv.html?highlight=csv%20writer#csv.reader">csv.reader</a> to parse these CSV files as&nbsp;<strong>no newlines</strong><strong>&nbsp;have been removed</strong>&nbsp;from developer comments.</p> <p>&nbsp;</p> <p><em><strong>88_million_developer_comments.zip</strong></em></p> <p>This zip file contains 135 CSV files; 3 per language. CSV names are formatted <em>&lt;code&gt;_&lt;language&gt;.csv</em>, with <em>&lt;language&gt;</em> being the name of the primary language and <em>&lt;code&gt;</em> being one of <em>co</em> (commits), <em>is</em> (issues), or <em>pr</em> (pull requests).</p> <p>Languages included are:&nbsp;ABAP, Assembly, C, C# (C-Sharp), C++ (C-PlusPlus), Clojure, COBOL, CoffeeScript,&nbsp;CSS, Dart, D, DM, Elixir, Fortran, F# (F-Sharp), Go, Groovy, HTML, Java, JavaScript, Julia, Kotlin, Lisp, Lua, MATLAB, Nim, Objective-C, Pascal, Perl, PHP, PowerShell, Prolog, Python, R, Ruby, Rust, Scala, Scheme, Scratch, Shell, Swift, TSQL, TypeScript, VBScript, and VHDL.</p> <p>Details on the columns in each CSV file are described in the provided <em>README.md</em>.</p> <p>&nbsp;</p> <p><em><strong>Detailed_Breakdown.ods</strong></em></p> <p>This spreadsheet contains specific details on how many repositories, commits, issues, pull requests, and comments are included in <em>88_million_developer_comments.zip</em>.</p> <p>&nbsp;</p> <p><em><strong>Note On Completeness</strong></em></p> <p>We make no guarantee that <em><strong>every</strong></em> commit, issue, and/or pull request for each repository is included in this dataset. Due to the nature of the GraphQL API and data decoding difficulties, sometimes a query failed and that data is not included here.</p> <p>&nbsp;</p> <p><em><strong>Versioning</strong></em></p> <ul> <li><strong>v1.1:</strong> The original corpus had duplicate header rows in the CSV files. This has been fixed.</li> <li><strong>v1.0:</strong> Original corpus.</li> </ul> <p>&nbsp;</p> <p><em><strong>Contact</strong></em></p> <p>Please contact Benjamin S. Meyers (<a href="mailto:bsm9339@rit.edu">email</a>) with questions about this data and its collection.</p> <p>&nbsp;</p> <p><em><strong>Acknowledgments</strong></em></p> <ul> <li>Collection of this data has been sponsored in part by the National Science Foundation grant 1922169, and by a Department of Defense DARPA SBIR program (grant 140D63-19-C-0018).</li> <li>This data was collected using the compute resources from the <a href="https://www.rit.edu/researchcomputing/">Research Computing</a> department&nbsp;at the Rochester Institute of Technology. <a href="https://doi.org/10.34788/0S3G-QD15">doi:10.34788/0S3G-QD15</a></li> </ul>

ShareScore

40/100

Overall dataset sharing score

Score breakdown

These five areas show where the dataset supports — or may limit — practical reuse.

Stewardship
8
Harmonization
4
Access
16
Reuse readiness
8
Engagement
4

Topics

Dataset details

Repository
zenodo
Published
October 24, 2021
License
cc-by-4.0
DOI
10.5281/zenodo.5596536

Scored with release 0.9.0

ScienceDex for faculty

Your academic CV and NIH biosketch

Make your CV in your school's format and your NIH biosketch from it, then download both as Word files.