A look at a 2026 Wikimania Hackathon Project – Wikidata Bias Detector

The Wikimania conference has a multitude of pathways to engage with Wikimedia projects and fellow Wikimedians. One such pathway is the hackathon. Having run at the Wikimania conference for years, the Hackathon provides an opportunity for the Wikimedia community to collaborate on technical projects and solve problems together.

In this blog post, we look at one of the projects worked on at the 2026 Wikimania Hackathon, a Wikidata Bias Detector. This project was created by Clair Kronk (Superraptor123), and as the name says, the goal is to look at bias in Wikidata, specifically for humans, to direct future work. This started as an early concept exploration and, by the end of the Hackathon, has moved to a beta deployment. Clair has set up a Github for this project, https://github.com/Superraptor/debias-wikidata, where you can follow the progress.

Some of the early results are very interesting and may stimulate discussion in the community. Clair would like to stress the Beta nature of the tool package, and there are still some issues that need to be overcome. One is being able to query the whole Wikidata set around humans. Currently, the dataset is too large and the queries time out. As such, the results are based on samples of 500 responses. Clair would be very appreciative if anyone could help overcome this issue (please contact Superraptor123).

With those caveats, the early results are highlighting biases found by others. For example, in the samples, men are 1.6x overrepresented (17.7% for women vs 82.3% for men). This is similar to what has been found by the Women in Red project, where only 20% of bibliographic articles on English Wikipedia are on women.

Another interesting finding is where people are born – urban vs rural locals. In the sample dataset, urban birth locations accounted for 85% of data. While according to the United Nation’s estimates – based on satellite data – 57% of people live in urban locations. There are caveats to the UN data, it is based on satellite data so the actual number may be different. However, we are likely talking in the range of 55-60% so the 85% is still well outside the error range. As Clair explained, this might not necessarily be a bias in the Wikidata, there could be cultural/economic/other reasons that people in urban locations dominate in areas of notability in general and Wikidata is just reflecting that.

This theme of outside issues affecting the data is highlighted in another finding: Iran, Egypt, China, India, and Brazil account for 18% of the world’s population but only 0.3% of people in the dataset. However, China bans access to Wikimedia projects for its population and given the local nature of who/what is added to Wikidata, this missing data, in some cases, is the result of local laws/censorship.

These early results highlight that there are biases in Wikidata when it comes to humans, but the reasons for these biases are varied and, in some cases, will need further research to understand why they are occurring. Clair’s toolkit is a great first step in examining Wikidata and its biases that can guide future work in Wikidata and it is great to see something so useful to come out of the Wikimania Hackathon.