A Search Engine for Algorithmic Fairness Datasets Alessandro Fabris 0 1 Fabio Giachelle 1 Alberto Piva 1 Gianmaria Silvello 1 Gian Antonio Susto 1 Max Planck Institute for Security and Privacy University of Padova

Algorithmic equity is a key desideratum for systems embedded in a diverse society producing data with embedded patterns of discrimination. This data is leveraged in algorithmic fairness research with the aim of studying the root causes of undesirable discrimination and developing methods to overcome them. Data documentation is central in supporting discoverability and correct use of existing resources. Documentation debt causes suboptimal data usage, with a negative impact on data-driven research and practice. This work introduces a search engine for algorithmic fairness datasets, describing its scope, functionality, and envisioned use cases, calling for inputs and collaboration within the community for the long-term maintenance and exploitation of this resource.

 eol>Algorithmic Fairness Fairness Datasets Documentation Debt Information Access 1. Introduction

Algorithmic Fairness is a scholarly field aimed at ensuring equity in algorithmic decision making [ 1 ], with dedicated measures [ 2, 3 ], algorithms [ 4, 5 ], and auditing procedures [ 6, 7 ]. Many of the key findings in this field have been data-driven [ 8, 9 ]. Therefore, the quality of datasets employed in research and practice are central to the validity of experiments and to the generalization of results in algorithmic fairness. Downstream efects of data issues triggered by poor practice that undervalues data quality are both common and avoidable [ 10 ]. Noisy, inaccurate, or otherwise non-representative data inevitably afect the reliability and utility of ifndings [ 11, 12 ]. Algorithmic fairness, as a whole, stands to gain from improvements in its prevalent data practices.

Recent work has shown that algorithmic fairness articles frequently use “of-the-shelf” datasets [13]. Fabris et al. [14] call into question the suitability of these benchmark datasets in algorithmic fairness, documenting over 200 alternative datasets that have been employed in the ifeld. In this work, we develop a search engine that makes the documentation of Fabris et al. [14] readily available and searchable.1

 2. Key Functionality 3. Use Cases and Applications

The WebApp we developed supports research and practice in algorithmic fairness and critical data studies in several ways. Below are the main use cases we envision.

1. Enabling task-driven and domain-driven search for principled dataset selection. Researchers and practitioners with a specific research angle may use our WebApp to find the most suited datasets for their needs. 2. Supporting multi-dataset studies in identifying relevant resources; for example, studies of how race and gender are encoded in datasets can use our tool to select datasets with the sensitive attributes of interest. 3. Directing data audits and critical data studies towards important resources; for example, datasets used in many research articles are pivotal for the community and deserve deeper scrutiny. 4. Highlighting under-explored domains or tasks, where new contributions, such as algorithms and datasets, can have a larger impact.

 4. Discussion and Call for Contributions

Documentation debt causes suboptimal data usage and negatively afects data-driven research [ 10, 15 ]. Our WebApp aims to empower the algorithmic fairness community, enabling principled approaches to select datasets for research, development, and critical data studies. Our longterm goal is to support easily accessible, up-to-date search along relevant axes. Updating and maintaining this resource with new datasets will certainly be a challenge.

We call on the algorithmic fairness community, the key stakeholders of this work, to contribute with guidance and collaboration, to help shape and maintain this resource. [13] B. Laufer, S. Jain, A. F. Cooper, J. Kleinberg, H. Heidari, Four years of facct: A reflexive, mixed-methods analysis of research contributions, shortcomings, and future prospects, in: 2022 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’22, Association for Computing Machinery, New York, NY, USA, 2022, p. 401–426. URL: https: //doi.org/10.1145/3531146.3533107. doi:10.1145/3531146.3533107. [14] A. Fabris, S. Messina, G. Silvello, G. A. Susto, Algorithmic fairness datasets: the story so far,

Data Mining and Knowledge Discovery (2022). doi:10.1007/s10618-022-00854-z. [15] E. M. Bender, T. Gebru, A. McMillan-Major, S. Shmitchell, On the dangers of stochastic parrots: Can language models be too big?, FAccT ’21, Association for Computing Machinery, New York, NY, USA, 2021, p. 610–623. URL: https://doi.org/10.1145/3442188.3445922. doi:10. 1145/3442188.3445922.

 [1] S. Barocas , M. Hardt , A. Narayanan , Fairness and Machine Learning, fairmlbook .org, 2019 . http://www.fairmlbook.org. [2] A. Castelnovo , R. Crupi , G. Greco , D. Regoli , I. G. Penco , A. C. Cosentini , A clarification of the nuances in the fairness metrics landscape , Scientific Reports 12 ( 2022 ) 4209 . [3] A. Fabris , G. Silvello, G. A. Susto , A. J. Biega , Pairwise fairness in ranking as a dissatisfaction measure , in: Proceedings of the Sixteenth ACM International Conference on Web Search and Data Mining , WSDM '23, Association for Computing Machinery, New York, NY, USA, 2023 , p. 931 - 939 . URL: https://doi.org/10.1145/3539597.3570459. doi: 10 .1145/3539597. 3570459. [4] M. Hardt , E. Price , N. Srebro , Equality of opportunity in supervised learning , in: Proc. of the 29th Annual Conference on Neural Information Processing Systems (NIPS 2016 ), Barcelona, ES , 2016 , pp. 3323 - 3331 . [5] S. C. Geyik , S. Ambler , K. Kenthapadi , Fairness-aware ranking in search & recommendation systems with application to linkedin talent search , in: Proceedings of the 25th acm sigkdd international conference on knowledge discovery & data mining , 2019 , pp. 2221 - 2231 . [6] A. Fabris , A. Mishler , S. Gottardi , M. Carletti , M. Daicampi , G. A. Susto , G. Silvello, Algorithmic audit of italian car insurance: Evidence of unfairness in access and pricing , in: Proceedings of the 2021 AAAI/ACM Conference on AI , Ethics , and Society, AIES '21, Association for Computing Machinery, New York, NY, USA, 2021 , p. 458 - 468 . URL: https://doi.org/10.1145/3461702.3462569. doi: 10 .1145/3461702.3462569. [7] A. Fabris , A. Esuli , A. Moreo , F. Sebastiani , Measuring fairness under unawareness of sensitive attributes: A quantification-based approach , Journal of Artificial Intelligence Research 76 ( 2023 ) 1117 - 1180 . [8] S. A. Friedler , C. Scheidegger , S. Venkatasubramanian , S. Choudhary , E. P. Hamilton , D. Roth , A comparative study of fairness-enhancing interventions in machine learning , in: Proceedings of the Conference on Fairness, Accountability, and Transparency , FAT* ' 19 , Association for Computing Machinery, New York, NY, USA, 2019 , p. 329 - 338 . URL: https://doi.org/10.1145/3287560.3287589. doi: 10 .1145/3287560.3287589. [9] J. Buolamwini , T. Gebru, Gender shades: Intersectional accuracy disparities in commercial gender classification , in: Conference on fairness, accountability and transparency, PMLR , 2018 , pp. 77 - 91 . [10] N. Sambasivan , S. Kapania , H. Highfill , D. Akrong , P. Paritosh , L. M. Aroyo , “ everyone wants to do the model work, not the data work”: Data cascades in high-stakes ai , in: proceedings of the 2021 CHI Conference on Human Factors in Computing Systems , 2021 , pp. 1 - 15 . [11] M. F. Kilkenny , K. M. Robinson , Data quality:“garbage in-garbage out” , volume 47 , SAGE Publications Sage UK : London, England, 2018 , pp. 103 - 105 . [12] J. Hullman , S. Kapoor , P. Nanayakkara , A. Gelman , A. Narayanan , The worst of both worlds: A comparative analysis of errors in learning from data in psychology and machine learning , in: Proceedings of the 2022 AAAI/ACM Conference on AI , Ethics , and Society, AIES '22, Association for Computing Machinery, New York, NY, USA, 2022 , p. 335 - 348 . URL: https://doi.org/10.1145/3514094.3534196. doi: 10 .1145/3514094.3534196.