Fairness Measures

Datasets and software for detecting algorithmic discrimination







Code (on GitHub) 🔗



Fairness Ranking Algorithms

It is crucial that the decision-making algorithms are fair, that they do not transmit or amplify the already existing biases in today’s society. Researchers in the field of fairness algorithms have provided various possible algorithms. We will cite a few examples here.

Fair Top-k Ranking Algorithm

Zehlike et al provided an algorithmic solution to the Fair Top-k Ranking problem over a single binary type attribute. With respect to the group fairness criteria, they provided a greedy algorithm which selects the best k candidates from a large pool while ensuring maximal utility.

Designing Fair Ranking Schemes

Asudeh et al. have focused on designing fair scoring functions by assigning a weighted sum of numeric attribute values to each item.

Fairness of Exposure in Rankings

Singh and Joachims have proposed a general framework of fairness constraints that employs probabilistic rankings and linear programming, allowing the computation of the utility- maximising ranking.

Ranking with Fairness Constraints

Celis et al. provided a linear time approximation algorithm of the ranking problem in ethical data processing.


The above described algorithms can be found in following literature:

Zehlike, Meike, et al. “Fa* ir: A fair top-k ranking algorithm.” Proceedings of the 2017 ACM on Conference on Information and Knowledge Management. doi:10.1145/3132847.3132938. bibtex

Asudeh, Abolfazl, et al. “Designing Fair Ranking Schemes.”arXiv preprint arXiv:1712.09752 (2017). bibtex

Singh, Ashudeep, and Thorsten Joachims. “Fairness of Exposure in Rankings.” arXiv preprint arXiv:1802.07281 (2018).bibtex

Celis, L. Elisa, Damian Straszak, and Nisheeth K. Vishnoi. “Ranking with fairness constraints.” arXiv preprint arXiv:1704.06840 (2017).bibtex