Session | ||
SB04 - SIG Service2: Machine learning in action
| ||
Presentations | ||
Cold start to improve market thickness on online advertising platforms: data-driven algorithms and field experiments 1University of Illinois at Urbana Champaign, United States of America; 2Washington University in St. Louis; 3Arizona State University; 4New York University Shanghai To solve the cold start problem on advertising platforms, we build a data-driven optimization model that captures the essential trade-off between short-term revenue and long-term market thickness on the platform, and propose a bandit algorithm to solve the problem. We also demonstrate the effectiveness of our algorithm via a novel two-sided randomized field experiment, and show our algorithm increases the cold start success rate by 62% and boosts the platform’s overall market thickness by 3.1%. Synthetically Controlled Bandits 1Operations Research Center, Massachusetts Institute of Technology, United States of America; 2Graduate School of Business, Columbia University, United States of America; 3Sloan School of Management, Massachusetts Institute of Technology, United States of America; 4Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, United States of America We present a dynamic experimental design for settings where the experimental units are coarse (e.g. to mitigate interference). `Region-split' experiments on online platforms are one such setting. Our design, dubbed Synthetically Controlled Thompson Sampling (SCTS), minimizes the cost (i.e. regret) associated with experimentation at no meaningful loss to inferential ability. We provide theoretical guarantees and experiments highlighting the merits of SCTS relative to fixed and switchback designs. |