Fair Ranking Under Uncertainty

Venkataraman, Brinda

Student Work

Fair Ranking Under Uncertainty

Public Deposited

In today's modern world, digital media and algorithms increasingly control what is seen by users online. This can have serious ramifications when the rankings could perpetuate bias against legally protected groups. These ranking methods must therefore aim for fairness towards such protected groups. However, when necessary demographic information is not available, institutions must look to outside inference algorithms for the information. In our research, we analyze the effectiveness of in-processing and post-processing fair ranking methods under uncertain demographic information. Our results show that using inference algorithms that are less accurate with respect to the protected group with a fair ranking method produces more fair rankings. However, we recommend future researchers to spend more time tuning the parameters of the fair ranking methods. Additional work was done by one team member to fulfill the requirements of her double major.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator