Reproducibility study of "FairLISA: Fair User Modeling with Limited Sensitive Attributes Information"
Abstract: This is a reproducibility study of the paper "FairLISA: Fair User Modeling with Limited Sensitive Attributes Information" by Zhang et al. (2023). It proposes a method of increasing fairness in user modeling tasks, by filtering out sensitive information from user embeddings. In contrast to other fairness aware methods, FairLISA is designed for filtering data with both known and unknown sensitive attributes. In this paper we explain the method from the paper, the claims about the effectiveness of the method, and our process of attempting to recreate said claims. We test the reproducibility of their original claims, test the generalisability of their method, and provide our implementation of the FairLISA method so further research can be done. We conclude that none of the claims of the original paper are fully reproducible in a reasonable amount of time. Some of the claims were able to be partially reproduced, and we detail those results.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Aurélien_Bellet1
Submission Number: 2257
Loading