Professor of Computer Science Thorsten Joachims, along with fellow researchers Marco Morik (a visiting undergraduate intern), Ashudeep Singh (Ph.D. candidate), and Jessica Hong '20, won the Best Paper Award at the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval for "Controlling Fairness and Bias in Dynamic Learning-to-Rank."
In accounting for the significance of the inquiry, Joachims, who is on the Executive Committee for the Center for Data Science for Enterprise and Society, notes that "ranking systems determine how easily each item can be discovered, greatly influencing which products get sold and which information gets seen—to just name a few examples." In describing the group's findings, Joachim emphasizes that "the paper provides a method to enforce fairness guarantees to the items that are being ranked, making it possible to implement fairness policies even though the underlying system is still learning the ranking function." Read the article.
Rankings are the primary interface through which many online platforms match users to items (e.g. news, products, music, video). In these two-sided markets, not only the users draw utility from the rankings, but the rankings also determine the utility (e.g. exposure, revenue) for the item providers (e.g. publishers, sellers, artists, studios). It has already been noted that myopically optimizing utility to the users—as done by virtually all learning-to-rank algorithms—can be unfair to the item providers. We, therefore, present a learning-to-rank approach for explicitly enforcing merit- based fairness guarantees to groups of items (e.g. articles by the same publisher, tracks by the same artist). In particular, we propose a learning algorithm that ensures notions of amortized group fairness, while simultaneously learning the ranking function from implicit feedback data. The algorithm takes the form of a controller that integrates unbiased estimators for both fairness and utility, dynamically adapting both as more data becomes available. In addition to its rigorous theoretical foundation and convergence guarantees, we find empirically that the algorithm is highly practical and robust.