Rui Qin

A Budget Optimization Framework for Search Advertisements Across Markets

Budget optimization is one of the primary decision-making issues faced by advertisers in search auctions. A quality budget optimization strategy can significantly improve the effectiveness of search advertising campaigns, thus helping advertisers to succeed in the fierce competition of online marketing. This paper investigates budget optimization problems in search advertisements and proposes a novel hierarchical budget optimization framework (BOF), with consideration of the entire life cycle of advertising campaigns. Then, we formulated our BOF framework, made some mathematical analysis on some desirable properties, and presented an effective solution algorithm. Moreover, we established a simple but illustrative instantiation of our BOF framework which can help advertisers to allocate and adjust the budget of search advertising campaigns. Our BOF framework provides an open testbed environment for various strategies of budget allocation and adjustment across search advertising markets. With field reports and logs from real-world search advertising campaigns, we designed some experiments to evaluate the effectiveness of our BOF framework and instantiated strategies. Experimental results are quite promising, where our BOF framework and instantiated strategies perform better than two baseline budget strategies commonly used in practical advertising campaigns.

Dynamic dual adjustment of daily budgets and bids in sponsored search auctions

As a form of targeted advertising, sponsored search auctions attract advertisers bidding for a limited number of slots in paid online listings. Sponsored search markets usually change rapidly over time, which requires advertisers to adjust their advertising strategies in a timely manner according to market dynamics. In this research, we argue that both the bid price and the advertiser (claimed) daily budget should be dynamically changed at a fine granularity (e.g., within a day) for an effective advertising strategy. By doing so, we can avoid wasting money on early ineffective clicks and seize better advertising opportunities in the future. We formulate the problem of dual adjusting (claimed) daily budget and bid price as a continuous state - discrete action decision process in the continuous reinforcement learning (CRL) framework. We fit the CRL approach to our decision scenarios by considering market dynamics and features of sponsored search auctions. We conduct experiments on a real-world dataset collected from campaigns conducted by an e-commerce advertiser on a major Chinese search engine to evaluate our dual adjustment strategy. Experimental results show that our strategy outperforms two state-of-the-art baseline strategies and illustrate the effect of adjusting either (claimed) daily budget or bid price in advertising.

Budget Planning for Coupled Campaigns in Sponsored Search Auctions

Budget-related decisions in sponsored search auctions are recognized as a structured decision problem rather than a simple constraint. Budget planning over several coupled campaigns (e.g., substitution and complementarity) remains a challenging but important task for advertisers. In this paper, we propose a dynamic multicampaign budget planning approach using optimal control techniques, with consideration of the substitution relationship between advertising campaigns. A three-dimensional measure of substitution relationships between campaigns is presented, namely, the overlapping degree in terms of campaign contents, promotional periods, and target regions. We also study some desirable properties and possible solutions to our budget model. Computational simulations and experiments are conducted to evaluate our model using real-world data from practical campaigns in sponsored search auctions. Experimental results show that (1) our approach outperforms the baseline strategy that is commonly used in practice; (2) coupled campaigns with a higher overlapping degree in between reduce the optimal total budget level, then reduce the optimal payoff, and reach the budgeting cap earlier than those with a less overlapping degree; and (3) the advertising effort can be seriously weakened by ignoring the degree of overlapping between campaigns.

Budget Strategy in Uncertain Environments of Search Auctions: A Preliminary Investigation

How to rationally allocate the limited advertising budget is a critical issue in sponsored search auctions. There are plenty of uncertainties in the mapping from the budget into the advertising performance. This paper presented some preliminary efforts to deal with uncertainties in search marketing environments, following principles of a hierarchical budget optimization framework (BOF). We proposed a stochastic, risk-constrained budget strategy, by considering a random factor of clicks per unit cost to capture a kind of uncertainty at the campaign level. Uncertainties of random factors at the campaign level lead to risk at the market/system level. We also proved its theoretical soundness through analyzing some desirable properties. Some computational experiments were made to evaluate our proposed budget strategy with real-word data collected from reports and logs of search advertising campaigns. Experimental results illustrated that our strategy outperforms two baseline strategies. We also noticed that 1) the risk tolerance has great influences on the determination of optimal budget solutions; 2) the higher risk tolerance leads to more expected revenues.

Exploring the optimal granularity for market segmentation in RTB advertising via computational experiment approach

Real Time Bidding (RTB) is a novel business model of online computational advertising, developing rapidly with the integration of Internet economy and big data analysis. It evolves the business logic of online ad-delivery from buying ad-impressions in websites or ad slots to directly buying the best-matched target audiences, and thus can help advertisers achieve the precision marketing. As a critical part of RTB advertising markets, Demand Side Platforms (DSPs) play a central role in matching advertisers with their target audiences via cookie-based data analysis and market segmentation, and their segmentation strategies (especially the choice of granularity) have key influences in improving the effectiveness and efficiency of RTB advertising markets. Based on a mathematical programming approach, this paper studied DSPs strategies for market segmentation, and established a selection model of the granularity for segmenting RTB advertising markets. With the computational experiment approach, we designed three experimental scenarios to validate our proposed model, and the experimental results show that: 1) market segmentation has the potential of improving the total revenue of all the advertisers; 2) with the increasing refinement of the market segmentation granularity, the total revenue has a tendency of a rise first and followed by a decline; 3) the optimal granularity of market segmentation will be significantly influenced by the number of advertisers on the DSP, but less influenced by the number of ad requests. Our findings show the crucial role of market segmentation on the RTB advertising effect, and indicate that the DSPs should adjust their market segmentation strategies according to their total number of advertisers. Our findings also highlight the importance of advertisers as well as the characteristics of the target audiences to DSPs market segmentation decisions.

Research on the frequency capping issue in RTB advertising: A computational experiment approach

Real time bidding (RTB) is emerged with the rapid development and integration of Internet and big data, and it has become the most important business model for online computational advertising. In RTB-based advertising markets, Demand Side Platforms (DSPs) aim to help the advertisers buy ad impressions matched with their target audiences. Due to the existence of discount rate, the advertising effect may be diminished when displaying the advertisements multiple times to the same target audience. As such, frequency capping is widely considered as a crucial issue faced by most advertisers. In this paper, we mainly consider the frequency capping problems in RTB advertising markets, and establish a two-stage optimization model for advertisers and DSPs. Utilizing the computational experiment approach, we design two experiments to validate our model. The experimental results show that under different discount rates, the optimal frequency caps are different. Moreover, when considering all the discount rates, there exists an optimal frequency cap, at which the expected maximum revenue can be obtained in the long run.

Analyzing the Segmentation Granularity of RTB Advertising Markets: A Computational Experiment Approach

Real Time Bidding (RTB) is an emerging business model of online computational advertising with the rise of Internet and big data. It can help advertisers achieve the precision marketing through evolving the traditional business logic from buying ad-impressions to directly buying the matched target audiences. As an important part of RTB markets, Demand Side Platforms (DSPs) play a critical role in matching advertisers with their target audiences via market segmentation, and their segmentation strategies (especially the choice of granularity) have key influences in improving the efficiency of RTB markets. This paper studied DSPs’ strategies for market segmentation, and established a selection model of the granularity for segmenting RTB markets. We proposed to validate our model using a computational experiment approach, and the experimental results show that the market segmentation granularity has the potential of improving both the total revenue of all the advertisers and the expected revenue for each advertiser.

A stochastic model for budget distribution over time in search advertisements

In search advertisements, advertisers have to seek for an effective allocation strategy to distribute the limited budget over a series of sequential temporal slots (e.g., days). However, advertisers usually have no sufficient knowledge to determine the optimal budget for each temporal slot, because there exist much uncertainty in search advertising markets. In this paper, we present a stochastic model for budget distribution over a series of sequential temporal slots under a finite time horizon, assuming that the best budget is a random variable. We study some properties and feasible solutions for our model, taking the best budget being characterized by either normal distribution or uniform distribution, respectively. Furthermore, we also make some experiments to evaluate our model and identify strategies with the real-world data collected from practical advertising campaigns. Experimental results show that a) our strategies outperform the baseline strategy that is commonly used in practice; b) the optimal budget is more likely to be normally distributed than uniformly distributed.

Boundary value problems for stochastic budget distribution in search advertisements

In sponsored search auctions, advertisers have to distribute the budget to a series of temporal slots in order to maximize the expected revenue. There exists a budget demand for each temporal slot, which can not be known exactly by the advertiser due to some uncertainties in the search marketing environments. The estimation of the value range of budget demand in a temporal slot seriously affects the advertising performance. In this paper we study the effect of the value range on the revenue and conduct some experiments to validate our model and identified properties with the real-world data collected from practical advertising campaigns. Experimental results show that, under a certain condition, (a) the higher estimation of the upper bound and the lower bound might increase the expected revenue, and (b) the expected revenue is positively proportional to the mean value of the value range and is negatively proportional to the size.