Document Type

Dissertation

Date of Award

Spring 2017

Degree Name

Doctor of Philosophy in Information Systems - (Ph.D.)

Department

Information Systems

First Advisor

Yi Chen

Second Advisor

Cristian Borcea

Third Advisor

Vincent Oria

Fourth Advisor

Ellen Thomas

Fifth Advisor

Nitish Korula

Abstract

As a massive industry, display advertising delivers advertisers’ marketing messages to attract customers through graphic banners on webpages. Display advertising is also the most essential revenue source of online publishers. Currently, advertisers are charged by user response or ad serving. However, recent studies show that users barely click or convert display ads. Moreover, about half of the ads are actually never seen by users. In this case, advertisers cannot enhance their brand awareness and increase return on investment. Publishers also lose much revenue. Therefore, the ad pricing standards are shifting to a new model: ad impressions are paid if they are viewable, not just being responded to or served. The Media Ratings Council’s standard for a viewable display impression is a minimum of 50% of pixels in view for a minimum of one second. To implement viewable impressions as pricing currency, ad viewability should be accurately predicted. Ad viewability prediction can improve the performance of guaranteed ad delivery, real-time bidding, as well as recommender systems.

This research is the first to address this important problem of ad viewability prediction. Inspired by the standard definition of viewability, this study proposes to solve the problem from two angles: 1) scrolling behavior and 2) dwell time. In the first phase, ad viewability is predicted by estimating the probability that a user will scroll to the page depth where an ad is located in a specific page view. Two novel probabilistic latent class models (PLC) are proposed. The first PLC model computes constant use and page memberships offline, while the second PLC model computes dynamic memberships in real-time. In the second phase, ad viewability is predicted by estimating the probability that the page depth will be in-view for certain seconds. Machine learning models based on Factorization Machines (FM) and Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) are proposed to predict the viewability of any given page depth in a specific page view. The experiments show that the proposed algorithms significantly outperform the comparison systems.

Share

COinS