Human-centered Machine Learning for Online Marketplaces

Understanding and influencing human decisions is paramount to today’s online marketplaces. To achieve the desired objective, the online marketplaces must learn how to efficiently influence human decisions and accurately understand human decision behavior in practice.  In this talk, I will draw from a human-centered view and use a data-driven approach to study the above problems. First, I will talk about how to design efficient data-driven algorithms to utilize the information to influence human decisions in a dynamic uncertain environment, and moreover, I will showcase recent findings illustrating how prices could also be carefully designed to achieve desired outcome, especially when humans are influenced by reference effects. Next, I will present behavior studies to understand human decision behavior in practice and show how the proposed behavior model affects the desired information design. In closing, I outline some future directions for human-centered machine learning in designing future online marketplaces, towards a vision of building efficient and human-friendly economic systems.