AdKDD2014: A Dynamic Pricing Model for Unifying Programmatic Guarantee and Real-Time Bidding in Display Advertising
There are two major ways of selling impressions in display advertising. They are either sold in spot through auction mechanisms or in advance via guaranteed contracts. The former has achieved a significant automation via real-time bidding (RTB); however, the latter is still mainly done over the counter through direct sales. This paper proposes a mathematical model that allocates and prices the future impressions between real-time auctions and guaranteed contracts. Under conventional economic assumptions, our model shows that the two ways can be seamless combined programmatically and the publisher’s revenue can be maximized via price discrimination and optimal allocation. We consider advertisers are risk-averse, and they would be willing to purchase guaranteed impressions if the total costs are less than their private values. We also consider that an advertiser’s purchase behavior can be affected by both the guaranteed price and the time interval between the purchase time and the impression delivery date. Our solution suggests an optimal percentage of future impressions to sell in advance and provides an explicit formula to calculate at what prices to sell. We find that the optimal guaranteed prices are dynamic and are non-decreasing over time. We evaluate our method with RTB datasets and find that the model adopts different strategies in allocation and pricing according to the level of competition. From the experiments we find that, in a less competitive market, lower prices of the guaranteed contracts will encourage the purchase in advance and the revenue gain is mainly contributed by the increased competition in future RTB. In a highly competitive market, advertisers are more willing to purchase the guaranteed con- tracts and thus higher prices are expected. The revenue gain is largely contributed by the guaranteed selling.
In this paper we study bid optimisation for real-time bid- ding (RTB) based display advertising. RTB allows adver- tisers to bid on a display ad impression in real time when it is being generated. It goes beyond contextual advertising by motivating the bidding focused on user data and it is different from the sponsored search auction where the bid price is associated with keywords. For the demand side, a fundamental technical challenge is to automate the bid- ding process based on the budget, the campaign objective and various information gathered in runtime and in history. In this paper, the programmatic bidding is cast as a functional optimisation problem. Under certain dependency assumptions, we derive simple bidding functions that can be calculated in real time; our finding shows that the optimal bid has a non-linear relationship with the impression level evaluation such as the click-through rate and the conversion rate, which are estimated in real time from the impression level features. This is different from previous work that is mainly focused on a linear bidding function. Our mathematical derivation suggests that optimal bidding strategies should try to bid more impressions rather than focus on a small set of high valued impressions because ac- cording to the current RTB market data, compared to the higher evaluated impressions, the lower evaluated ones are more cost effective and the chances of winning them are relatively higher. Aside from the theoretical insights, offline experiments on a real dataset and online experiments on a production RTB system verify the effectiveness of our proposed optimal bidding strategies and the functional optimisation framework.
the paper is downloadable at here.
In this paper we analyse the reserve price optimisation problem and report the first empirical study in the Real-Time Bidding (RTB) context. A reserve price is the minimum that the publisher would accept from bidders in auctions and it is one of the most important yield management tools. In the second price auction it could uplift revenue by charging winners the reserve price instead of the second highest bids. Comparing with other popular forms (sponsored search, contextual advertising, etc.), this problem in the RTB context has unique features, mainly because bidders have to submit a bid for each individual impression and using user data is greatly encouraged, as well as many practical constraints (budget, campaigns’ life time, etc.) and irrational decisions. These make inferring the private values very difficult thus make it interesting and necessary to examine commonly adopted algorithms in the real-world. Moreover, in RTB an advertiser is facing nearly unlimited supply which naturally encourages spending less on placements of high cost. This could imply the loss of bid volume over time if an aggressive reserve price is in place, which requires analysis on bidders’ attrition. In this paper we empirically study commonly adopted algorithms at placement level and propose variations. Especially we analyse challenge of fitting private values, which are foundation of the optimal auction theory, in RTB practice based on real-world data. Then, we represent our findings on RTB auctions and discuss their possible impact. Finally, we report a large scale online experiment in a production platform. The results suggest the game theory based OneShot policy performed the best and the superiority is significant in most cases. The analysis on bidders’ attrition show good chances for publishers to active this optimisation.
a copy can be downloaded here.
We are involved in the first global RTB bidding algorithm competition, organised by iPinYou. Our Phd Students Weinan Zhang and Shuai Yuan have won the third season offline competition. The result can be found here. In this season, data scientists from eight countries, including China, USA, UK, India and France, have participated in the competition and made over 2000 submissions. The IPinYou global RTB (Real-Time Bidding) algorithm competition kicks off on April 1, 2013. The purpose of this competition is to further improve the performance of DSP bidding algorithm, stimulate the interest of research and development of DSP bidding algorithms in the whole data science research community, and speed up the growth of RTB-enabled display advertising ecosystem. In the end, iPinYou expects this competition will help advertising technology companies serve our advertisers better. There are three Milestone Prizes and one Grand Prize (RMB 1,000,000) for the DSP bidding competition.
Online advertising is now one of the fastest advancing areas in IT industry. In display and mobile advertising, the most significant development in recent years is the growth of Real-Time Bidding (RTB), which allows selling and buying online display advertising in real-time one ad impression at a time. Since then, RTB has fundamentally changed the landscape of the digital media market by scaling the buying process across a large number of available inventories. It also encourages behaviour (re-)targeting, and makes a significant shift toward buying focused on user data, rather than contextual data. A report from IDC shows that in 2011, global RTB based display ad spend increased by 237% compared to 2010, with the U.S.’s $2.2 billion RTB display spend leading the way. The market share of RTB-based spending of all display ad spending will grow from 10% in 2011 to 27% in 2016, and its share of all indirect spending will grow from 28% to 78%.
Scientifically, the further demand for automation, integration and optimization in RTB brings new research opportunities in the CIKM fields. For instance, the much enhanced flexibility of allowing advertisers and agencies to maximize impact of budgets by more optimised buys based on their own or 3rd party (user) data makes the online advertising market a step closer to the financial markets, where unification and interconnection are strongly promoted. The unification and interconnections across webpages, advertisers, and users require significant research on knowledge management, data mining, information retrieval, behaviour targeting and their links to game theory, economics and optimization.
Despite its rapid growth and huge potential, many aspects of RTB remain unknown to the research community for a variety of reasons. In this tutorial, teamed up with presenters from both the industry and academia, we aim to bring the insightful knowledge from the real-world systems, to bridge the gaps between industry and academia, and to provide an overview of the fundamental infrastructure, algorithms, and technical and research challenges of the new frontier of computational advertising.
Financial Methods in Computational Advertising
Computational Advertising has recently emerged as a new scientific sub-discipline, bridging the gap among the areas such as information retrieval, data mining, machine learning, economics, and game theory. In this tutorial, I shall present a number of challenging issues by analogy with financial markets. The key vision is that display opportunities are regarded as raw material “commodities” similar to petroleum and natural gas – for a particular ad campaign, the effectiveness (quality) of a display opportunity shouldn’t rely on where it is brought and whom it belongs, but it should depend on how good it will benefit the campaign (e.g., the underlying web users’ satisfactions or respond rates). With this vision in mind, I will go through the recently emerged real-time advertising, aka Real-Time Bidding (RTB), and provide the first empirical study of RTB on an operational ad exchange. We show that RTB, though suffering its own issue, has the potential of facilitating a unified and interconnected ad marketplace, making it one step closer to the properties in financial markets. At the latter part of this talk, I will talk about Programmatic Premium, i.e., a counterpart to RTB to make display opportunities in future time accessible. For that, I will present a new type of ad contracts, ad options, which have the right, but no obligation to purchase ads. With the option contracts, advertisers have increased certainty about their campaign costs, while publishers could raise the advertisers’ loyalty. I show that our proposed pricing model for the ad option is closely related to a special exotic option in finance that contains multiple underlying assets (multi-keywords) and is also multi-exercisable (multi-clicks). Experimental results on real advertising data verify our pricing model and demonstrate that advertising options can benefit both advertisers and search engines.
In this paper, we study collaborative filtering (CF) in an interactive setting, in which a recommender system continuously recommends items to individual users and receives interactive feedback. Whilst users enjoy sequential recommendations, the recommendation predictions are constantly refined using up-to-date feedback on the recommended items. Bringing the interactive mechanism back to CF process is fundamental because the ultimate goal for a recommender system is about the discovery of interesting items for individual users and yet users’ personal preferences and contexts evolve over time during the interactions with the system. This requires us not to distinguish between the stages of collecting information to construct the user profile and making recommendations, but to seamlessly integrate these stages together during the interactive process, with the goal of maximizing the overall recommendation accuracy throughout the interactions. This mechanism naturally addresses the cold-start problem as any user can immediately receive sequential recommendations without providing ratings beforehand. We formulate Interactive CF with the probabilistic matrix factorization (PMF) framework, and leverage several exploitation-exploration algorithms to select items, including the empirical Thompson sampling and upper-confidence-bound-based algorithms. We conduct our experiment on cold-start users as well as warm-start users with drifting taste. Results show that the proposed methods have significant improvements over several strong baselines for the MovieLens, EachMovie and Netflix datasets.
In sponsored search, advertising slots are usually sold by a search engine to an advertiser through an auction mechanism in which advertisers bid on keywords. In theory, an auction mechanism encourages the advertisers to truthfully bid for keywords. However, keyword auctions have a number of problems including: (i) volatility in revenue, (ii) uncertainty in the bidding and charged prices for advertisers’ keywords, and (iii) weak brand loyalty between the advertiser and the search engine. To address these issues, we study the possibility of creating a special option contract that alleviates these problems. In our proposal, an advertiser purchases an option in advance from a search engine by paying an upfront fee, known as the option price. The advertiser then has the right, but no obligation, to then purchase specific keywords for a fixed cost- per-click (CPC) for a specified number of clicks in a specified period of time. Hence, the advertiser has increased certainty in sponsored search while the search engine could raise the customers’ loyalty. The proposed option contract can be used in conjunction with traditional keyword auctions. As such, the option price and corresponding fixed CPC price must be set such that there is no arbitrage opportunity. In this paper, we discuss an option pricing model tailored to sponsored search that deals with spot CPCs of targeted keywords in a generalized second price (GSP) auction. We show that the pricing model for keywords is closely related to a special exotic option in finance that contains multiple underlying assets (multi-keywords) and is also multi-exercisable (multi-clicks). Experimental results on real advertising data verify our pricing model and demonstrate that advertising options can benefit both advertisers and search engines.
Personalization techniques have been widely adopted in many recommender systems. However, experiments on real-word datasets show that for some users in certain contexts, personalized recommendations do not necessarily perform better than recommendations that rely purely on popularity. Broadly, this can be interpreted by the fact that the parameters of a personalization model are usually estimated from sparse data; the resulting personalized prediction, despite of its low bias, is often volatile. In this paper, we study the problem further by investigating into the ranking of recommendation lists. From a risk management and portfolio retrieval perspective, there is no difference between the popularity-based and the personalized ranking as both of the recommendation outputs can be represented as the trade-off between expected relevance (reward) and associated uncertainty (risk). Through our analysis, we discover common scenarios and provide a technique to predict whether personalization will fail. Besides the theoretical understanding, our experimental results show that the resulting switch algorithm, which decides whether or not to personalize, outperforms the mainstream recommendation algorithms.