Document Type
Dissertation
Date of Award
12-31-2020
Degree Name
Doctor of Philosophy in Business Data Science - (Ph.D.)
Department
School of Management
First Advisor
Yi Chen
Second Advisor
Cristian Borcea
Third Advisor
Dantong Yu
Fourth Advisor
Michael A. Ehrlich
Fifth Advisor
Jorge Eduardo Fresneda Fernandez
Abstract
The revenue of online display advertising is expected to reach 389 billion U.S. dollars globally in 2021. However, the fast increase in ad-blocker usage impacts publishers' ad revenue and the ad-supported free web. Faced with this problem, many online publishers choose either to cooperate with ad-blocker software companies to show acceptable ads or to build a wall that requires users to whitelist (i.e., turn off ad-blockers) for content access. This dissertation aims to deepen the understanding of existing counter-ad-blocking strategies and further explore how to address ad-blocking challenges through a data science approach.
The dissertation starts with investigating and identifying factors influencing ad-blocker usage. A field-study experiment is conducted on the website of a large online publisher, in which the ad-blocker users are randomly assigned into a treatment group, which receives the wall strategy, and a control group, which receives the acceptable ads strategy. The experiment shows that the wall strategy has an overall negative impact on user engagement. However, it has no statistically significant effect on highly engaged users as their engagement is not changed by the wall strategy. On the other hand, it has a big impact on lowly engaged users, who have no loyalty to the site. The study also shows that the number of users visiting the site decreases over time, but the whitelisting ratio increases over time as the remaining users have relatively high loyalty and high engagement. Our study further provides managerial insights for publishers: it is recommended to allow new users to bypass the wall strategy in order to strengthen their attachment to the website.
After studying the effect of existing strategies on ad-blocking, the dissertation investigates new counter-ad-blocking strategies. A novel personalized dynamic counter-ad-blocking strategy is proposed based on whitelist prediction. If a user is predicted to whitelist, they are directed to the wall strategy. Otherwise, they are directed to the acceptable ad strategy. A deep learning model, named Deep Ad-Block Whitelist Network (DAWN), is proposed to predict whether a user would whitelist a page when facing with the wall. DAWN represents the history of page visits and user actions on pages using an attention mechanism. DAWN also considers multitask learning on both whitelist prediction and dwell time prediction to enhance the model learning ability on parameter training. This is the first study in the literature to develop personalized counter-ad-blocking strategies. Experiments with data from a large publisher demonstrate the effectiveness of the proposed personalized counter-ad-blocking strategy over existing strategies.
Last, it is known that recommending relevant articles can provide better user experience and improve user engagement, which may convince ad-blocker users to whitelist. However, existing page recommendation models do not work well in the presence of new privacy regulations, such as General Data Protection Regulation. Under these regulations, users may refuse to share data such as their browsing history or demographics with the websites. This study proposes Fed4Rec, a privacy-preserving framework for page recommendation based on federated learning and model-agnostic meta-learning, which trains machine learning models jointly on data collected from both public users, who share data with the server, and private users, who do not share data with the server. Experiments using a publicly available dataset from a large news portal demonstrate that Fed4Rec performs better than the comparison systems, and it is especially beneficial when the percentage of public users is low.
Recommended Citation
Zhao, Shuai, "Addressing ad-blocking challenges in online publishing: a data science approach" (2020). Dissertations. 1911.
https://digitalcommons.njit.edu/dissertations/1911
Included in
Artificial Intelligence and Robotics Commons, Business Administration, Management, and Operations Commons
