Machine Learning Optimization in Computational Advertising—A Systematic Literature Review

Abstract

Most of the advertisements seen on PC and mobile devices these days are delivered by machines. Training them to understand user behaviors and the surrounding context is still an ongoing challenge that advertising networks are facing. This study reviewed the literature related to machine learning optimization in that so-called computational advertising. The purpose is to find the applications, techniques, and challenges of using machine learning algorithms in optimizing this new type of advertising. Following the PRISMA framework, hundreds of manuscripts were initially identified in the current literature databases by their titles, abstracts, and keywords. They were then carefully screened with our inclusion and exclusion criteria to select qualified ones for full paper reviews. By the end, this study found that the current algorithm literature is mostly focusing on either estimating clicks or deciding the bid price. Little research has been done for optimizing conversions and the bid cost. More importantly, machine learning algorithms are mostly designed for the demand side of the advertisement serving process, leaving the supply side largely under-optimized. The advertising creative process and customer-retention activities still require extensive human efforts. Those challenges urgently call for more research in this emerging area, which is positioned at the crossroad of business and technology.