Real-Time Data Analytics and Optimization for Computational Advertising
Online advertising has built a market of hundreds of billions of dollars and still continues to grow. With well developed techniques in big data storage, data mining and analytics, online advertising is able to reach targeted audiences e ctively. Real- time bidding refers to the buying and sellin...
| Other Authors: | |
|---|---|
| Format: | Others |
| Language: | English |
| Published: |
Florida Atlantic University
|
| Subjects: | |
| Online Access: | http://purl.flvc.org/fau/fd/FA00004940 http://purl.flvc.org/fau/fd/FA00004940 |
| Summary: | Online advertising has built a market of hundreds of billions of dollars and still
continues to grow. With well developed techniques in big data storage, data mining
and analytics, online advertising is able to reach targeted audiences e ctively. Real-
time bidding refers to the buying and selling of online ad impressions through ad
inventory auctions which occur in the time it takes a webpage to load. How to de-
termine the bidding price and how to allocate the budget of advertising is the key to
successful ad campaigns. Both of these aspects are fundamental to most campaign
optimizations and we will introduce both of them in this thesis. For bidding price
determination, we improved the estimation of CTR (Click Through Rate) (one of the
most important factors of determining the bidding price) by using a re ned hierar-
chical tree structure for the estimation. The result of the experiment and the A/B
test showed our proposal can provide stable improvement. For budget allocation, we
introduce SCO (Single Campaign Optimization) and CCO (Cross Campaign Opti-
mization). SCO has been applied by our commercial partner while CCO needs more
research. We will rst introduce the methods of SCO and then give our proposal
about CCO. We modeled CCO as a LP (Linear Programming) problem as well as designed an e ective procedure to implement optimal impressions distribution. Our
simulation showed our proposal can signi cantly increase global Gross Pro t (GP). === Includes bibliography. === Thesis (M.S.)--Florida Atlantic University, 2017. === FAU Electronic Theses and Dissertations Collection |
|---|