Fair Ranking in Information Retrieval
Jean-Michel Renders, Naver Labs Europe, Meylan, France
Talk at the Search Engines course, A.Y. 2021/2022 ([ Ссылка ])
MD in Computer Engineering ([ Ссылка ])
MD in Data Science ([ Ссылка ])
University of Padua, Italy
Abstract:
With the growing consideration of ethical aspects and algorithmic accountability in Information Retrieval (IR), ensuring fair ranking when delivering search results is now becoming an important direction of research. This talk aims at introducing the main concepts of fairness in ranking systems, as well as the current families of approaches to solve the associated problems. We will start by describing the potential sources of unfairness in IR. We will then explain what fundamentally differentiates the “fair ranking” task from the “fair classification” task, which is a well-studied problem in Machine Learning. In particular, we will summarize and compare the multiple definitions of fairness. This will be followed by a survey of recent methods, where we will compare them both in the objectives they try to optimize (typically a mix of utility and a particular definition of fairness) and the technical frameworks adopted to solve the problem. We will end the talk by addressing the challenge of multi-sided fairness and by sketching future promising directions for Fair Ranking in IR.
Bio:
Dr. Jean-Michel RENDERS is Research Fellow at Naver Labs Europe (France). Previously, he was leading the Search and Recommendation team of Naver Labs Europe. His main research domain focuses in exploiting unstructured data (text, speech, images, …) from multiple, heterogeneous sources in order to separate useful signals from noise in a large variety of applications, often with real-time and scalability constraints. He is leading/has lead research projects that cover various technical domains including Natural Language Processing, Search & Recommendation Engines, Fairness and Bias in Machine Learning, and privacy-preserving IA.
