Three methods for fair ranking of multiple protected items

对多个受保护项目进行公平排序的三种方法

阅读:1

Abstract

Three approaches to fair ranking in retrieval systems are compared in this paper: mPFR, which is based on the theory of preferences and eigensystems; cRR, which is a simple' 'round robin" method; and mMLP, which is based on linear programming. In order to increase fairness without sacrificing retrieval effectiveness, the techniques post-process the rankings that a retrieval system sends back to users. The findings demonstrate that when it comes to protecting elements, mPFR and cRR accomplish the same level of effectiveness and fairness. Despite being computationally more costly than the latter, the former's mathematical architecture enables the ranking of reordering techniques at various levels of complexity, while mMLP might not be practical for datasets that are too big. Therefore, the choice between these methods often hinges on the specific use case and dataset size, where trade-offs between computational efficiency and desired fairness come into play. Future research could explore optimizing these techniques further to enhance their applicability across diverse scenarios, ensuring that both fairness and effectiveness are maintained.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。