Cold Start on Online Advertising Platforms: Data-Driven Algorithms and Field Experiments

A Scalable Recommendation Engine for New Users and Items

“…Cold start is a common challenge in the digital environment. Proposed approaches to address the cold-start problem span contexts such as ad bidding algorithms for new ads (Ye et al, 2020), customer relationship management for new users (Padilla and Ascarza, 2021), personalized website design for new visitors (Liberali and Ferecatu, Forthcoming), and mostly close to our context, recommendation systems with limited historical data on user-item interactions for new users or new items (e.g., Gonzalez Camacho and Alves-Souza, 2018;Hu et al, 2022;Zheng et al, 2018;Gardete and Santos, 2020). Gardete and Santos (2020) (Kumar et al, 2020;Wei et al, 2017;McInerney et al, 2018;Wang et al, 2017;Zhang et al, 2020;Hu et al, 2019;Guo et al, 2020;Gupta et al, 2020;Strub et al, 2016), network information (Gonzalez Camacho and Alves-Souza, 2018), or similarity measures that link new users (items) with existing users (items) (Bobadilla et al 2012;Zheng et al 2018).…”

“…Our method also builds on a large literature applying online experiments to learn about uncertainties in marketing (Schwartz et al, 2017;Misra et al, 2019;Liberali and Ferecatu, Forthcoming;Ye et al, 2020;Gardete and Santos, 2020;Aramayo et al, Forthcoming), operations management (Bertsimas and Mersereau, 2007;Bernstein et al, 2019;Johari et al, 2021), and computer science (Li et al, 2010;Gomez-Uribe and Hunt, 2015;Wu, 2018;Zhou, 2015;Silva et al, 2022;Gangan et al, 2021). Specifically, we focus on Multiarmed Bandit, a classic reinforcement learning problem (Katehakis and Veinott, 1987).…”

Bid Landscape Forecasting and Cold Start Problem With Transformers

“…This problem arises when a new advertisement enters the auction system, and there is insufficient historical data to train a reliable model for bid prediction. Previous approaches [7], [8] have tried to manage this by using simple average bid values or adapting data from similar ads.…”