Our Thoughts

Data Science
Journal Club

Copilot data on 2021 Data Science Journal Club birthday

Our Data Science team is always tapped in to the latest thought leadership in our industry and beyond. Browse our collection of reviewed journals and see our thoughts on each.

Inefficiencies in Digital Advertising Markets

Brett R Gordon, Kinshuk Jerath, Zsolt Katona, Sridhar Narayanan, Jiwoong Shin, Kenneth C Wilbur

This paper surveys four inefficiencies common in digital advertising: issues in measuring the real effect of ads on consumer behavior; organizational inefficiency in and between companies in the advertising channel; ad blocking and its ramifications on addressable audiences; and fraud in digital advertising. Authors discuss academic literature relevant to each topic, recent developments, and the ways that these problems' manifestations in digital advertising are different than those of other industries in which similar inefficiencies may be present.

Bid Optimization by Multivariable Control in Display Advertising

Xun Yang, Yasong Li, Hao Wang, Di Wu, Qing Tan, Jian Xu, Kun Gai

Consider the problem of finding optimal bids for display advertising campaigns. Typically, advertisers task DSPs with gaining as much value as possible, while adhering to certain budget constraints. Beyond this, advertisers often require that the campaign meet certain Key Performance Indicators (KPIs). In this case, the authors consider the problem of maximizing the quantity of conversion events while meeting a cost-per-click KPI requirement. The authors use linear programming to approach this problem and derive an optimal strategy.

A Multi-Armed Bandit Model Selection for Cold-Start User Recommendation

Crícia Felício, Klérisson Paixão, Celia Barcelos, Philippe Preux

Explore the "cold start" problem for user recommendation algorithms: how do we deal with making recommendations for users about whom we have no prior information? Although various algorithms exist to attempt to make these inferences, it is typically unclear from the start which ones will be most effective. This paper discusses a Multi-Armed Bandit approach to balancing different recommendation methods — enabling an explore-exploit approach, in which various algorithms are simultaneously tested against real users. The most effective algorithms get used increasingly often, thus maximizing overall performance.

Combining Powers of Two Predictors in Optimizing Real-Time Bidding Strategy under Constrained Budget

Chi-Chun Lin, Kun-Ta Chuang, Wush Chi-Hsuan Wu, Ming-Syan Chen

This paper applies the "knapsack problem" to ad bidding: if you're filling a knapsack with items that may be valuable, then the more space any one item takes up, the more confident you should be that it is likely to have real value. So in the same way, authors designed a method to predict the winning price of an ad auction. This is to go beyond the idea that more expensive ad inventory should just have a higher chance of clicks or conversions, as the authors argue that the more expensive inventory is, the more confident we should be in the accuracy of our predictions in the first place.