Wikipedia and Wikidata are socio-technical systems driven by collaborative communities, open content, and open-source infrastructure. Some of the open-source software development around them involves code-sharing sites like GitHub. Analyzing GitHub for repositories related to Wikipedia and Wikidata can thus provide insights into multiple dimensions of the development of Wikipedia and Wikidata tools. We plan to do such an analysis, and in order to test our workflows for doing that, we ran a preliminary study based on a sample of 1000 GitHub repositories each for Wikidata and Wikipedia. We are preregistering our workflows here as a transparent basis for documenting and reporting on the full analysis later. The kinds of insights we expect based on the preliminary data about open-source GitHub repositories related to Wikidata and Wikipedia are as follows: (i) statistical information about these repositories; (ii) computational information, e.g. in terms of the programming languages used; (iii) demographic information about the contributors to such open-source projects; (iv) legal information about the choice of licenses; (v) linguistic information about the natural language used in the context of these repositories; (vi) trends over time. In the process of applying these preliminary workflows to studying the full dataset of GitHub repositories related to Wikipedia and Wikidata, we hope to gain some additional insights into the community dynamics at play in volunteer software development around Wikimedia projects, as well as into the process and merits of preregistrations for studies of this kind. We welcome community feedback on this approach as well as suggestions on additional aspects to include into the full study, and collaborations on the actual implementation.
CEUR ceur-ws.org CEUR Workshop Proceedings
The field of Empirical Software Engineering has evolved around the idea and practice of studying
how programmers work, e.g. how they address a particular issue or use a given resource to
develop computer-based solutions [
This document represents a preregistration, outlining our approach to evaluating how programmers work on their development projects related to Wikipedia and Wikidata through the assessment of GitHub repositories about Wikidata and Wikipedia. Understanding the GitHub landscape of Wikipedia and Wikidata repositories is essential for optimizing collaboration, identifying programming trends, and ensuring the sustainability of these knowledge-sharing platforms. The primary research questions that guide our study include: • How does the GitHub community interact with repositories related to Wikipedia and
Wikidata? • What are the predominant programming languages used in GitHub repositories related to Wikipedia and Wikidata? • Who contributes the most to these GitHub repositories? • What licenses are commonly associated with open-source projects related to Wikipedia and Wikidata? • How is the number of repositories distributed by their year of creation and year of the last push? • What are the most common topics of the repositories related to Wikipedia and Wikidata? To address these questions, we will begin by explaining our approach to data collection and analysis based on a preliminary study using small-scale sample data and open-source code (Section 2). We will also highlight the importance of these research questions and their relevance in understanding the open-source landscape around Wikipedia and Wikidata. Following the data collection and analysis, we will provide preliminary results for this study based on the sample data and discuss them by contextualizing them with previous research findings on the matter (Section 3). Finally, we will draw conclusions on what the preliminary results mean for a larger-scale study and for potential future work targeting the open-source landscape around Wikipedia and Wikidata (Section 4). Each figure and table presented on the way will contain a brief comment on how we expect the methodology of the full study to compare to the preliminary methodology presented here, emphasizing the significance of addressing the research questions to contribute to the understanding of software development in the Wikimedia community.
For the initial study, we focus on GitHub, querying its search API via the dedicated Python
library PyGitHub [
The retrieved data involves multiple variables about the characteristics of repositories and their
activity, as shown in Table 1. We stored the dataset as an Excel spreadsheet using Pandas, a
Python Library for data analytics [
Of the 2000 GitHub repositories retrieved in total, some (18 for Wikipedia and 6 for Wikidata)
were empty. Since that does not provide information about software development, we eliminated
them in favour of a more meaningful analysis, leaving 1,976 non-blank GitHub repositories
related to either or both of Wikipedia or Wikidata. 982 of them are exclusively related to
Wikipedia, while 994 are related to Wikidata4. This distribution allows a fair comparison
4Wikidata-related repositories include 35 generic repositories also linked to Wikipedia. We disregard this fact for
the purpose of our preliminary study. For the full study, we expect this number to be higher, and plan to explore
the connections.
between the patterns in which Wikipedia and Wikidata are used in computer programming,
whereas the full study will mean a comparison based on unequal numbers (somewhat higher
than the 36.2k for Wikipedia versus 2.9k for Wikidata that were reported above). 78.3% of the
Wikidata-related repositories and 68.6% of the Wikipedia-related ones returned a value for the
Home Page URL, so presumably have a project website. This can be interpreted to mean that
the Wikidata-related repositories and to a lesser extent the Wikipedia-related repositories are
developed for practical use in deploying web tools and services [
When assessing the popularity and volume of the GitHub repositories, we found that the
average number of stars and forks for Wikipedia-related repositories significantly exceeds the
ones for Wikidata-related repositories, as shown in Figure 1. This is mainly due to the higher
popularity and longer age of Wikipedia. Wikipedia has been created in 2001, so it is 11 years
older than Wikidata [
5Live data: https://stats.wikimedia.org/#/wikidata.org and https://stats.wikimedia.org/#/en.wikipedia.org.
When examining the main programming language for GitHub repositories related to
Wikidata and Wikipedia, we found a similar profile about the choice of programming languages
to process Wikidata and Wikipedia, as shown in Figure 2, despite the fact that Wikipedia is
semi-structured, while Wikidata is a fully-structured open knowledge graph [
When retrieving the GitHub usernames of the accounts behind the repositories, we found the
repositories to be distributed among developers in a way that follows Lotka’s law [
The situation is diferent for Wikidata, where we observed Wikimedia volunteers who had
established more than three relevant repositories, including Tiago Lubiana (lubianat) and
Maxime Lathuilière (maxlath) at 13 and 12, respectively. Wikimedia researchers having a
close relationship with the Wikimedia Community are also visibly engaged in Wikidata
opensource development (e.g., Egon Willighagen [egonw], Andrew I. Su [sulab], and Jakob Voß
[nichtich]). The higher involvement of the Wikimedia Community in developing
Wikidatarelated projects can be explained in part by the more standardized format of Wikidata in the
form of triples, which makes it easier than Wikipedia to process automatically [
Among the information that have been retrieved using the GitHub API, there was data about license attribution based on the first line of every license file. These first lines of license files are processed by hand to attribute the right license name for every repository. The analysis of the licenses assigned to the GitHub repositories related to Wikidata and Wikipedia finds both to have a similar profile of license attribution, as shown in Figure 4. The apparent dominance of All Rights Reserved might come as a surprise but has to be understood such that those repositories did not have any license declared, in which case full copyright protection has to be assumed by default. Further eforts should be aimed at raising awareness within the Wikimedia technical community around the importance of using – and properly declaring – permissive licenses to allow the reuse and upgrade of their source codes for the good of the Wikimedia projects.
The most popular open license in Wikimedia open-source development is the MIT License,
followed by the GNU General Public License and the Apache License. These three licenses are
the main ones that have been used for years in open-source development on GitHub [36, 37].
The only surprising fact is that the rate of GNU General Public License use is higher in
Wikidatarelated repositories in the same way as the rate of MIT License is higher in Wikipedia-related
repositories. The reduced gap between GNU General Public License use and MIT License use in
Wikidata-related repositories can perhaps be explained by the fact that Wikidata is released
under the CC0 License [
Changes over time are dificult to assess with the limited dataset that we have, since we suspect that the sort order we used to sample GitHub repositories is influenced by at least some parameters that correlate with repository age, e.g. the number of stars, forks, commits, commiters, or trafic.
In particular, the sharp decline in recent years that is visible in Figure 5 might well be an artifact. This could have been introduced due to our sampling a limited amount of repositories in a non-random fashion, or it could be due to more subtle efects, such as the time that it takes before a GitHub-hosted repository is actually indexed by GitHub and included into its search index.
We thus refrain from interpreting the distribution given in Figure 5 based on the preliminary data, except for noting the existence of a few repositories that deal with Wikidata and that have been created before the inception of the Project in 2012. These projects are either preliminary sources for the initial development of Wikidata [31] or several Wikipedia-related development projects that have been adapted to support Wikidata as a resource like YAGO [40], an open knowledge graph initially derived from Wikipedia and WordNet [40].
Notwithstanding the suspected bias against new repositories, recent years feature strongly in terms of the year of last commit, and 2023 – despite being in its early stages at the time of sampling – came out on top. Once we have the full dataset, we expect the prominence of recent years in this plot to be even more pronounced, and comparisons to the temporal patterns of repository creation (as per Figure 5) might yield insights into community dynamics, sustainability and related matters. There could also be relationships between the license choice (cf. Figure 4) and commit trends, since more permissive licenses provide more avenues for engagement with a given repository [41].
The Description data obtained from GitHub for each repository (cf. Table 1 - Brief statements available for all the considered repositories) have been converted to lowercase, split by spaces, and stripped from stopwords and punctuation to identify the most common words provided to describe the GitHub repositories related to Wikidata and Wikipedia (cf. Fig. 7).
The analysis of the top words included in these descriptions revealed that the distribution of the words in the descriptions follows Zipf’s law [42].
Both sets of GitHub repositories mainly deal with applications for the processing of data of the two projects, as highlighted by the prominence of the word data in both. This involves the management of Wikipedia pages (articles, pages, text, article, and page) and Wikidata items and statements (knowledge, items, graph, and entities). This also includes the development of software ( code), software libraries ( library) – especially in python) – for the development of Application Programming Interfaces (APIs) and dump processing methods (api, wikibase, mediawiki, dumps, and dump) and the creation of tools (tool and tools) and projects (project). This is mostly done in the context of promoting the systematic use of APIs and dumps for the automatic enrichment and processing of Wikidata and Wikipedia [43].
Beyond this, there are several specific applications that are only applicable to Wikidata or
Wikipedia. Wikidata repositories emphasize projects related to the management of SPARQL
queries (sparql, queries, and query) or the creation of bots or scripts for the automatic enrichment
of the knowledge graph. This is closely linked to ongoing research projects for Wikidata related
to SPARQL query optimization and data augmentation based on external resources [
As for Wikipedia repositories, they show an interest in the development of methods for quicker and simpler data mining of the project (search and simple). This confirms that the development eforts around Wikipedia meet the long-term research eforts for the development of robust and more eficient data mining techniques for exploring Wikipedia [ 44].
This project lightly analyzed about 2000 repositories out of nearly 40,000 that exist on GitHub. Now that this project exists as a demonstration of how the analysis could work, we could gain higher precision in our findings and also expand to more easily identify exceptional cases if we continued to analyze all identified repositories. Furthermore, we could include GitLab or Codeberg repositories – which have a reputation among some Wikimedia developers for being more value-aligned than GitHub – or repositories from other platforms like Gitee that have other demographic biases. We could also take a look at the technical communication platforms of the Wikimedia Foundation such as Gerrit6 and Phabricator 7 [45] to further examine how
7https://phabricator.wikimedia.org. the Wikimedia technical community discusses the incremental development of collaborative projects, possibly including hardware-related ones like Internet-in-a-Box8.
This project explored which questions would be answerable with the data available in GitHub.
More insights could be gained from matching external datasets to this data (e.g. as per [
The Wikimedia community is highly engaged in the governance of Wikimedia projects. This engagement plays out in various ways, but for example, community forums exist in the Wikimedia platform where developers and users meet to discuss user challenges and technical possibilities. While the Wikimedia community will discuss this paper as they routinely do for all such reports, some people will undoubtedly ask further questions, and others may want to interpret the preliminary results from this paper to inform ongoing or planned tool development. We intended for this first analysis to be useful, but given the long-term budget planning of the Wikimedia Foundation for development, the global and large Wikimedia audience base, and the stakes of sustaining success as a nonprofit general information resource, scheduled reporting updates for development trends such as these would surely guide stakeholder decision making. One notable concern highlighted by our analysis is the prevalence of ”All Rights Reserved” licenses in repositories. This finding underscores the importance of educating developers on open science best practices and encouraging them to declare a license for their projects, even if they are not open-source in nature. This step can contribute to a more open and collaborative development ecosystem within Wikimedia and similar communities. This preregistration is only an early step into characterizing the development landscape. Correct collection and interpretation of development statistics such as those we explored here – especially considering how much of this is volunteer-organized with little central planning – could have significant returns of community engagement on the investment.
While Wikipedia and Wikidata have been success stories in many respects – including some software-related ones –, we suspect that volunteer developers of software around Wikipedia, Wikidata and other Wikimedia projects would still benefit from insights into the social dynamics of software development (within Wikimedia contexts as well as more generally), and so we invite feedback from anyone who might be a potential user of the results of the full study.
This research is funded by the Wikimedia Research Fund of Wikimedia Foundation (San Francisco, California, United States of America) through the Adapting Wikidata to support clinical practice using Data Science, Semantic Web and Machine Learning Project.9 Source code and data are made available under the MIT License at https://github.com/csisc/WikiGitHub.
9https://meta.wikimedia.org/wiki/Research:Adapting_Wikidata_to_support_clinical_practice_using_Data_Science, _Semantic_Web_and_Machine_Learning Conference on Global Software Engineering (ICGSE), IEEE, 2017. doi: 10.1109/icgse. 2017.11. [31] D. Vrandečić, L. Pintscher, M. Krötzsch, Wikidata: The Making Of, in: Companion Proceedings of the ACM Web Conference 2023, ACM, 2023. doi:10.1145/3543873.3585579. [32] Z. Wang, Y. Wang, D. Redmiles, Competence-confidence gap: a threat to female developers’ contribution on github, in: Proceedings of the 40th International Conference on Software Engineering: Software Engineering in Society, ACM, 2018. doi: 10.1145/3183428. 3183437. [33] B. Vasilescu, D. Posnett, B. Ray, M. G. van den Brand, A. Serebrenik, P. Devanbu, V. Filkov, Gender and Tenure Diversity in GitHub Teams, in: Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, ACM, 2015. doi:10.1145/2702123. 2702549. [34] J. Terrell, A. Kofink, J. Middleton, C. Rainear, E. Murphy-Hill, C. Parnin, J. Stallings, Gender diferences and bias in open source: pull request acceptance of women versus men, PeerJ Computer Science 3 (2017) e111. [35] S. Allison-Cassin, A. Armstrong, P. Ayers, T. Cramer, M. Custer, M. Lemus-Rojas, S. McCallum, M. Profitt, M. A. Puente, J. Ruttenberg, A. Stinson, ARL White Paper on Wikidata: Opportunities and Recommendations, Association of Research Libraries, Washington, DC, 2019. URL: https://www.arl.org/wp-content/uploads/2019/04/2019.04. 18-ARL-white-paper-on-Wikidata.pdf. [36] C. Vendome, G. Bavota, M. D. Penta, M. Linares-Vásquez, D. German, D. Poshyvanyk, License usage and changes: a large-scale study on GitHub, Empirical Software Engineering 22 (2016) 1537–1577. doi:10.1007/s10664-016-9438-4. [37] X. Wu, J.-Y. Wu, M.-H. Zhou, Z.-Q. Wang, L.-Y. Yang, Analysis of open source license selection for the GitHub programming community, 2020. doi:10.48550/ARXIV.2009.00981. [38] J. M. Heilman, E. Kemmann, M. Bonert, A. Chatterjee, B. Ragar, G. M. Beards, D. J. Iberri, M. Harvey, B. Thomas, W. Stomp, M. F. Martone, D. J. Lodge, A. Vondracek, J. F. de Wolf, C. Liber, S. C. Grover, T. J. Vickers, B. Meskó, M. R. Laurent, Wikipedia: A Key Tool for Global Public Health Promotion, Journal of Medical Internet Research 13 (2011) e14. doi:10.2196/jmir.1589. [39] G. Hagedorn, D. Mietchen, R. Morris, D. Agosti, L. Penev, W. Berendsohn, D. Hobern, Creative Commons licenses and the non-commercial condition: Implications for the re-use of biodiversity information, ZooKeys 150 (2011) 127–149. doi:10.3897/zookeys.150. 2189. [40] T. P. Tanon, G. Weikum, F. Suchanek, YAGO 4: A Reason-able Knowledge Base, in: The Semantic Web, Springer International Publishing, 2020, pp. 583–596. doi:10.1007/ 978-3-030-49461-2_34. [41] S. Rathee, A. Chobe, Open source growth and trends, in: Getting Started with Open Source
Technologies, Apress, 2022, pp. 149–169. doi:10.1007/978-1-4842-8127-7_8. [42] M. Newman, Power laws, Pareto distributions and Zipf's law, Contemporary Physics 46 (2005) 323–351. doi:10.1080/00107510500052444. [43] T. Steiner, Bots vs. Wikipedians, Anons vs. Logged-Ins (Redux), in: Proceedings of The International Symposium on Open Collaboration, ACM, 2014. doi:10.1145/2641580. 2641613. [44] Y. Wang, J. Zhang, Exploring topics related to data mining on Wikipedia, The Electronic
Library 35 (2017) 667–688. doi:10.1108/el-09-2016-0188. [45] W. Brown, How is the speed of code review afected by activity, usage and code quality?, 2023. doi:10.48550/ARXIV.2305.05770. [46] L. Rasberry, D. Mietchen, Scholia for Software, Research Ideas and Outcomes 8 (2022) e94771. doi:10.3897/RIO.8.E94771. [47] A.-M. Istrate, D. Li, D. Taraborelli, M. Torkar, B. Veytsman, I. Williams, A large dataset of software mentions in the biomedical literature (2022). doi: 10.48550/ARXIV.2209.00693.