Fairness and Bias of Machine Learning in Search and Ranking - Dissertation - 2024

Aug 5, 2026

Upload by Maeve

Recent advancements in information retrieval and machine learning improve ranking and search performance, yet training-data bias can cause unfair treatment of demographic groups and fuel systematic discrimination by race, gender, or geography. This research proposes frameworks to mitigate data bias and promote equitable representation and exposure in ranking and search outcomes. It introduces Meta-learning based Fair Ranking (MFR) and Meta Curriculum-based Fair Ranking (MCFR) using weighted loss and curriculum strategies, evaluates LLM text-ranking fairness, and studies selection bias in multi-stage recommendation with MUDA.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages143

Reading Time6 h 0 min

AccessFree to download

Introduction

Motivation

Overview

Contributions

Outline

Related Work

Fairness on Ranking

Meta-Learning on Fairness

Fairness in LLMs

Selection Bias

A Meta-learning Approach to Fair Ranking

Introduction

Meta-learning Based Fair Ranking

Experiments

Conclusion

A Unified Meta-learning Framework for Fair Ranking with Curriculum Learning

Introduction

Meta Curriculum-based Fair Ranking

Experiments

Santa Clara University Scholar Commons | Engineering Ph. D. Theses | Student Scholarship | | --- | --- | 5-2024 Fairness and Bias of Machine Learning in Search and Ranking Yuan Wang Follow this and additional works at: [https://scholarcommons.scu.edu/eng_phd_theses](https://scholarcommons.scu.edu/eng_phd_theses) Fairness and Bias of Machine Learning in Search and Ranking by Yuan Wang Dissertation Submitted in Partial Fulﬁllment of the Requirements for the Degree of Doctor of Philosophy in Computer Science & Engineering in the School of Engineering at Santa Clara University, 2024 Santa Clara, California Dedicated to my family iii Acknowledgements First and foremost, I extend my deepest gratitude to my advisor, Professor Yi Fang, whose unwavering guidance and support have been instrumental to my doctoral journey. Professor Fang not only believed in my potential but also supported me with patience, kindness, and an unmatched dedication to excellence. His mentorship transcended academic instruction, oﬀering personal support and invaluable life lessons that have shaped me both as a scholar and as an individual. He teaches me to keep trying, to be curious, and to work hard. I want to keep doing these things in my job too. Learning from him has been a really special chance for me, and I’m so thankful for it. I would like to thank my doctoral committee consisting of Prof. Zhiqiang Tao, Prof. David Anastasiu, Prof. Sean Choi, and Prof. Haibing Lu for their time and suggestions to make my thesis better. I would like to thank my lab mates Travis Ebesu, Xuyang Wu, Zhiyuan Peng, and Suthee Chaidaroon, who supported me through diﬀerent parts of this journey. Lastly, I would like to thank my family. My parents and my brother have given me endless love and support throughout this entire journey. I also want to thank my ﬁancée, Yijia, for her love and patience. Fairness and Bias of Machine Learning in Search and Ranking Yuan Wang Department of Computer Science & Engineering Santa Clara University Santa Clara, California 2024 ABSTRACT Recent advancements in Information Retrieval (IR) and machine learning have significantly improved ranking and search system performance. However, these data-driven approaches often suﬀer from inherent biases present in training datasets, leading to unfair treatment of certain demographic groups and contributing to systematic discrimination based on race, gender, or geographic location. This research aims to address the fairness and bias issue in ranking and search systems by proposing innovative frameworks that mitigate data bias and ensure equitable representation and exposure across diverse groups. We introduce two novel frameworks: the Meta-learning based Fair Ranking (MFR) model and the Meta Curriculum-based Fair Ranking (MCFR) framework, both designed to alleviate dataset bias through automatically-weighted loss functions and curriculum learning strategies, respectively. These approaches utilize meta-learning to adjust ranking loss, focusing particularly on improving the fairness metrics for minority groups while maintaining competitive ranking performance. Additionally, we conduct an empirical evaluation of Large Language Models (LLMs) in text-ranking tasks, revealing biases in handling queries and documents related to binary protected attributes. Our analysis oﬀers a benchmark for assessing LLMs’ fairness and highlights the necessity for equitable representation in search outcomes. Furthermore, we explore the challenge of data selection bias in multi-stage recommendation systems, particularly in online advertising contexts like Pinterest’s multi-cascade ads ranking system. Through comprehensive experiments, we assess various state-ofthe-art methods, and our ﬁndings demonstrate the eﬀectiveness of a modiﬁed version of unsupervised domain adaptation (MUDA) in mitigating selection bias. Collectively, our work contributes to the development of

FAQs

What problem does the dissertation address in search and ranking systems?

It addresses inherent bias in training data that can lead to unfair treatment of demographic groups and discrimination in ranking and search outcomes.

What are the two main fairness frameworks proposed by the research?

The dissertation proposes Meta-learning based Fair Ranking (MFR) and Meta Curriculum-based Fair Ranking (MCFR). MFR uses automatically weighted loss via meta-learning, while MCFR incorporates curriculum learning strategies.

How does the dissertation evaluate bias beyond traditional ranking models?

It conducts empirical evaluation of large language models in text-ranking tasks, analyzing biases in handling queries and documents related to protected attributes.

Aug 5, 2026

Upload by Maeve