Towards Global Optimization in Display Advertising by Integrating Multimedia Metrics with Real-Time Bidding

Towards Global Optimization in Display Advertising by Integrating Multimedia Metrics with Real-Time Bidding

Xiang Chen School of Computing
National University of Singapore, Singapore, 117417

Real-time bidding (RTB) has become a new norm in display advertising where a publisher uses auction models to sell online user’s page view to advertisers. In RTB, the ad with the highest bid price will be displayed to the user. This ad displaying process is biased towards the publisher. In fact, the benefits of the advertiser and the user have been rarely discussed. Towards the global optimization, we argue that all stakeholders’ benefits should be considered. To this end, we propose a novel computation framework where multimedia techniques and auction theory are integrated. This doctoral research mainly focus on 1) figuring out the multimedia metrics that affect the effectiveness of online advertising; 2) integrating the discovered metrics into the RTB framework. We have presented some preliminary results and discussed the future directions.

Display advertising; real-time bidding; multimedia metrics; trade-offs optimization
journalyear: 2017copyright: rightsretainedconference: MM’17; ; October 23–27, 2017, Mountain View, CA, USA.price: doi: ISBN 978-1-4503-4906-2/17/10

1. Introduction

The explosive growth of multimedia data on Internet has created huge opportunities for online advertising. According to the latest Interactive Advertising Bureau’s annual report (iab_2016), the second-half revenue totaled $39.8 billion in 2016, which is an increase of $7.7 billion from the second-half revenue of 2015. Despite the increasing growth of revenue, the effectiveness of online advertising remains debatable. On the one hand, the user gets annoyed by the improper ads. Marketing data (hubspot_annoy_ad) shows that: the average click-through rate of display ads across all formats and placement is 0.06%; young adults tend to ignore online banner ads; and ad blocking has grown by 41% globally in the year 2015. The unpleasant ad experience further influences their site visits, and their perception of the displayed ad. On the other hand, the advertiser gets harmed by ineffective ad delivery. In terms of cost-per-impression pricing mechanism, the advertiser has to pay for every impression. Due to user’s low engagement towards the displayed ad, a large amount of budget that the advertiser spends on online advertising is wasted.

The majority of online displayed ads are served through RTB. Compared to the RTB system that focuses on the publisher’s revenue (local maximization), we define the global optimization as the optimal trade-offs among all stakeholders. In this regard, the proper ad is not only the highest revenue from the economic point of view, but also the best fit with the context from the multimedia point of view. Fig. 1 illustrates how the multimedia metrics are incorporated with RTB. Our proposed framework consists of two stages: 1) the first stage uses the second-price (SP) auction model to deal with economic issues; 2) the second stage uses our proposed optimal re-ranking model to select the proper ad. Note that, the independent process of stage I ensures the equilibrium in ad auctions. The multimedia metrics take effect in stage II. Since they are exogenous, our proposed framework does not affect the advertiser’s bidding behavior.

Due to the complex nature of display advertising, we are facing the following challenges:

  • What metrics to select? User’s engagement (e.g., view, click) towards the displayed ad varies in different context. Understanding the context and user’s behavior becomes the key factor to improve the effectiveness of online advertising.

  • How to integrate the metrics? Online advertising is a multidisciplinary topic, which involves economics, multimedia and psychology. The metrics represent the stakeholders’ benefit from different domains. An easy but effective solution is the linear combination model. Thus, determining the optimal weights becomes the key issue.

The development of computer vision, multimedia analysis and machine learning techniques brings opportunities to solve these two challenges. Firstly, we select a set of multimedia metrics according to marketing data and consumer psychology literature. We then conduct a user-study experiment to demonstrate the effectiveness of the selected metrics. Secondly, we simulate a real-world advertising system based on an auction dataset and two multimedia datasets. By adding constrains on the changes of metrics when compared with existing system, we propose an optimization model to obtain the optimal weights. To the best of our knowledge, we are the first towards global optimization in display advertising by combining multimedia techniques and auction theory.

The rest of the paper is organized as follows. Section 2 reviews the related work. Section LABEL:sec:approach describes the proposed optimization framework. Section LABEL:sec:experiments presents our preliminary results. Section LABEL:sec:work_in_progress lists our works in progress. And section LABEL:sec:conclusion concludes the paper.

Figure 1. Overview of the proposed framework (chen2017RTBoptimizing).

2. Related work

From the economics perspective, researchers mainly focus on investigating the strategy of selling ad impressions. The generalized second-price (GSP) auction (Edelman_2007_2) and the Vickrey-Clark-Groves (VCG) auction (Parkes_2007) have been widely used on different platforms. As revenue is always the primary concern, the squashing parameter and reserve price have been discussed to increase the revenue of GSP auctions (Lahaie_2007; Thompson_2013). GSP auction model has been proved to be able to improve social welfare as well (Lahaie_2011).

From the multimedia perspective, researchers mainly focus on increasing user’s engagement towards displayed ads and meanwhile maintaining user’s online experience. Literature in marketing and consumer psychology has already shown that the contextual relevance between the content of the hosting webpage and the ads makes a large difference in their clickability (chatterjee2003modeling), and it also has a leading effect on user’s online experience (mccoy2007effects). According to the utilized information for contextual similarity matching, we can classify most existing contextual advertising systems into three categories: text-based advertising (li2010pagesense), visual-based advertising (sengamedu2007vadeo) and targeted advertising (wang2009argo). To achieve better contextual relevance, the multimodal approach has been proposed which considers both textual and visual information for video (mei2007videosense; guo2009adon) and image advertising (mei2012imagesense).

The works discussed so far only focus on the benefit of a specific stakeholder, regardless of the other stakeholders’ benefits. Trade-offs among multiple objectives or stakeholders have been discussed in several works. Likhodedov et al. (Likhodedov_2003) proposed a framework that linearly combines revenue and social welfare in a single-item auction. Bachrach et al. (Bachrach_2014) proposed a framework that linearly combines relevant, welfare and clicks in sponsored search. Liao et al. (liao2008adimage) combined revenue and relevance for video advertising.

Our goal is to foster a vigorous and healthy online advertising ecosystem. Different from most existing works that focus on local maximum value, we focus on the global optimization of the whole advertising ecosystem. Display advertising is an interplay among all stakeholders and each stakeholder’s ad experience is an important factor for advertising effectiveness. In the proposed framework, the publisher’s revenue will decrease in a short term in trade of improvements on the benefits of the advertiser and user. The increased ad experience will increase the demand of the advertiser’s advertising needs and the supply of the user’s webpage visits, which will boost the publisher’s revenue in a long run.

Figure 2. User’s engagement on displayed ad under different combination of multimedia metrics.
Variable \adl@mkpreamc—\@addtopreamble\@arstrut\@preamble Input for Stage II

Method Input Output
Publisher’s revenue Stage I Bid Payment normalized
Advertiser’s utility Stage I Bid, payment Utility normalized
Ad memorability MemNet (khosla2015understanding) Ad image Memorability score normalized score
Ad CTR Given by data CTR CTR normalized CTR
Contextual relevance TakeLab (vsaric2012takelab) Title, keywords, description Semantic similarity score normalized score
Ad saliency MBS (zhang2015minimum) Web page snapshot, ad image Saliency map and ratio normalized ratio
Table 1. Specifications of metric variables for Stage II.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description