Performance Measures and RTB Optimization

Performance Measures and RTB Optimization

Wenxue Huang (Guangzhou University, China), Yuanyi Pan (InferSystems Corp., Canada) and Jianhong Wu (York University, Canada)
Copyright: © 2014 |Pages: 9
DOI: 10.4018/978-1-4666-5202-6.ch165

Chapter Preview



The industrial applications of rare events targeting, besides online advertising, include banking and insurance modeling (King & Zeng, 2001; Pednault, Rosen & Apte, 2000; Vasu and Ravi, 2007), credit card fraud detection (Chan & Stolfo, 1998), network intrusion detection (Dreger, Feldmann, Paxson, & Sommer, 2004), and direct marketing (Ling, C., Ling, C.X. & Li, 1998) among others.

The industry of online advertising is growing fast. According to, US online ad spending will reach $62 billion in 2016 from $32 billion in 2011. Meanwhile, traditional media ad spending is descending or of no significant increasing. The newspaper and magazine ad speding was approxinately $36 billion and the TV ad spending was $60.7 billion in 2011, and these are expected to be $32.3 billion and $72 billion respectively in 2016.

The display advertisements, including online video, banner ads, rich media and sponsorships seems to have a prospective future, though the search advertising seems to have a bigger market for now, as shown in Figures 1 and 2 that illustrate the spending and the growth of spending in each format.

Figure 1.


Figure 2.

Spending growth


The demand side of the display advertising ecosystem, as shown in Figure 3, can be generalized as advertisers while the supply side being as the publishers. The advertisers have the demand to display ads and the publishers meet these demands. In the following, we adopt the abbreviation DSP and SSP for the platform for the demand side and the platform for the supply side respectively.

Figure 3.

Online display advertising ecosystem (Calic, 2010)


Key Terms in this Chapter

Advertiser: The demand side that has the need to display ads.

DSP: The demand side platform that collects display request from the advertisers and send them to SSP.

Gph: A replacement to the G -index that assign the topmost part of the lift table a significant weight.

Gip: Another replacement to the G -index that assign the weight of samples to each part of the lift table.

CPL: Clost per lead. It is an alias to CPA.

CPA: Cost per action. It needs further behavior after the click.

Gini Index (G-index): The overal accumulative counting of real positives in the lift table compared with that from a random prediction.

SSP: The supply side platform that bridges DSP and the publishers.

Confusion Matrix: A 2×2 matrix that indicates the true/false positive/negative in a binary prediction. It is the most applied statistics to evaluate the binary prediction result.

Lift Table: The counting of real positives in the target variable that is descendingly sorted by its numerical prediction and equally divided into certain parts.

RTB: Real time bidding. It is a trading system between DSP and SSP that allows DSPs to compete each other to get the display resource. SSP can then benefit from the competition since the competition increases the price to display ads.

CPC: Cost per click. It is another business model focusing on performance of the display as CPA and CPL.

Publisher: The supply side that has the resource to display ads.

Complete Chapter List

Search this Book: