PROBABILISTIC MODELS FOR AD VIEWABILITYPREDICTION ON THE WEB
ABSTRACT
Online display advertising has becomes a billion-dollar industry, and it keeps growing. Advertisers attempt to sendmarketing messages to attract potential customers via graphic banner ads on publishers’ webpages. Advertisers are charged for eachview of a page that delivers their display ads. However, recent studies have discovered that more than half of the ads are never shownon users’ screens due to insufficient scrolling. Thus, advertisers waste a great amount of money on these ads that do not bring anyreturn on investment. Given this situation, the Interactive Advertising Bureau calls for a shift toward charging by viewable impression,i.e., charge for ads that are viewed by users. With this new pricing model, it is helpful to predict the viewability of an ad. This paperproposes two probabilistic latent class models (PLC) that predict the viewability of any given scroll depth for a user-page pair. Using areal-life dataset from a large publisher, the experiments demonstrate that our models outperform comparison systems.
EXISTING SYSTEM:
Researchers have investigated scrolling behavior and viewabilityfor webpage usability evaluation. In, theauthors discover that users spend more time looking atinformation on the upper half of the page than the lowerhalf. Also, the distribution of the percentage of contentviewed by users follows a Gaussian-like distribution. Wediffer from these works in our main goal: viewability prediction.Existing work collects scrolling behaviorand uses it as an implicit indicator of user interests tomeasure webpage quality. In contrast, we design algorithmsto predict the scrolling behavior for any user-webpage pair.Several studies have attempted to predict user browsingbehavior, including click and dwelltime. The existing methods on click predictionare not applicable in our application. They rely heavily onside information (e.g., user profile, and users’ queries andtweets) in order to detect what the user is looking forand thereby suggest the items that are more likely to beclicked on. In our application, on the other hand, there is nosuch kind of explicit indicators of user information needsand detailed user profile. Wang et al. learn user’s clickbehavior from server logs in order to predict if a user willclick an ad shown for the query. The authors use featuresextracted from the queries to represent the user search intent.In our case, search queries, which can explicitly reflectuser interests, are not available.
PROPOSED SYSTEM:
The proposed methods have been experimentally comparedwith four systems: SVD, Cox Regression, LogisticRegression, and a deterministic method. The experimentsshow that our PLCs have better prediction performancethan the comparative systems. Also, PLCdyn has betteradaptability and less memory usage than PLC const.Our contributions are summarized as follows: 1) Wedefine the problem of viewability prediction for any pagedepth. 2) We propose two novel statistical models basedon PLC to predict the probability that a page depth willbe in-view. 3) We demonstrate experimentally using a reallifedataset that our two PLCs outperform three comparisonsystems. Compared with PLCconst, PLC dyn can savemore memory and better adapt to changes in user andwebpage characteristics
CONCLUSIONS
To the best of our knowledge, our research is the first tostudy the problem of predicting the viewability probabilityfor a given scroll depth and a user/webpage pair. Solvingthis issue can benefit online advertisers to allow them to investmore effectively in advertising and can benefit publishersto increase their revenue. We presented two PLC models,i.e., PLC with constant memberships and PLC with dynamicmemberships, that can predict the viewability for any givenscroll depth where an ad may be placed. The experimentalresults show that both PLC models have substantially betterprediction performance than the comparative systems. ThePLC with dynamic memberships can better adapt to theshift of user interests and webpage attractiveness and hasless memory consumption
REFERENCES
[1] I. Lunden, “Internet ad spend to reach $121b in 2014,”http://techcrunch.com/2014/04/07/internet-ad-spend-to-reach121b-in-2014-23-of-537b-total-ad-spend-ad-tech-gives-display-aboost-over-search/.
[2] Y. Chen and T. W. Yan, “Position-normalized click prediction insearch advertising,” in KDD’12, 2012, pp. 795–803.
[3] W. Zhang, S. Yuan, and J. Wang, “Optimal real-time bidding fordisplay advertising,” in ACM SIGKDD’14, 2014, pp. 1077–1086.
[4] W. Chen, D. He, T.-Y. Liu, T. Qin, Y. Tao, and L. Wang, “Generalizedsecond price auction with probabilistic broad match,” inACM EC’14, 2014, pp. 39–56.
[5] Google, “The importance of being seen,”http://think.storage.googleapis.com/docs/the-importanceof-being-seenstudy.pdf.
[6] M. Mareck, “Is online audience measurement coming of age?”Research World, vol. 2015, no. 51, pp. 16–19, 2015.
[7] S. Flosi, G. Fulgoni, and A. Vollman, “if an advertisement runsonline and no one sees it, is it still an ad?” Journal of AdvertisingResearch, 2013.
[8] H. Cheng, E. Manavoglu, Y. Cui, R. Zhang, and J. Mao, “Dynamicad layout revenue optimization for display advertising,” in Proceedingsof the Sixth International Workshop on Data Mining for OnlineAdvertising and Internet Economy, 2012, p. 9.
[9] H. Weinreich, H. Obendorf, E. Herder, and M. Mayer, “Not quitethe average: An empirical study of web use,” ACM TWEB, vol. 2,no. 1, p. 5, 2008.
[10] F. Manjoo, “You won’t finish this article,” Slate, 2013.
[11] E. Agichtein, E. Brill, and S. Dumais, “Improving web searchranking by incorporating user behavior information,” in ACMSIGIR’06, 2006, pp. 19–26