[论文解读] Bidding and Pricing in Budget and ROI Constrained Markets.
本文提出了一种针对在线广告市场中同时具有预算和投资回报率(ROI)约束的买家的基于阈值的出价框架,以及一种针对卖家的二分查找定价算法。在随机竞争环境下,该研究证明了买家和卖家算法的次线性遗憾,确保了约束满足,并实现了与事后最优策略相比的近似最优性能。
In online advertising markets, setting budget and return on investment (ROI) constraints are two prevalent ways to help advertisers (i.e. buyers) utilize limited monetary resources efficiently. In this work, we provide a holistic view of ROI and budget constrained markets. We first tackle the buyer's bidding problem subject to both budget and ROI constraints in repeated second-price auctions. We show that the optimal buyer hindsight policy admits a structure that suggests the buyer win all auctions during which her valuation-to-expenditure ratio is greater than some threshold. We further propose a threshold-based bidding framework that aims to mimic the hindsight bidding policy by learning its threshold. We show that when facing stochastic competition, our algorithm guarantees the satisfaction of both budget and ROI constraints and achieves sublinear regret compared to the optimal hindsight policy. Next, we study the seller's pricing problem against an ROI and budget constrained buyer. We establish that the seller's revenue function admits a bell-shaped structure, and then further propose a pricing algorithm that utilizes an episodic binary-search procedure to identify a revenue-optimal selling price. During each binary search episode, our pricing algorithm explores a particular price, allowing the buyer's learning algorithm to adapt and stabilize quickly. This, in turn, allows our seller algorithm to achieve sublinear regret against adaptive buyer algorithms that quickly react to price changes.
研究动机与目标
- 解决在重复的第二价格拍卖中,买家在同时具有预算和ROI约束下的高效资源配置挑战。
- 设计一种买家侧出价算法,以模仿最优事后出价策略,同时保证约束满足。
- 研究在存在ROI和预算约束买家的情况下,卖家的最优定价问题。
- 开发一种定价算法,通过分阶段二分查找识别出收益最优价格,并适应买家的学习动态。
- 在自适应、竞争性的环境中,实现买家和卖家算法的次线性遗憾。
提出的方法
- 买家的出价策略采用基于阈值的策略,当估值与支出之比超过学习到的阈值时即获胜。
- 买家的学习算法适应随机竞争,并在长时间内确保预算和ROI约束均被满足。
- 卖家采用分阶段二分查找过程来探索候选价格,从而实现对买家学习算法的稳定适应。
- 定价算法的每个阶段固定一个价格,并允许买家稳定,从而实现对收益的准确估计。
- 证明了卖家的收益函数具有钟形结构,可通过二分查找实现高效优化。
- 理论分析证明了买家和卖家算法相对于最优事后策略的次线性遗憾。
实验结果
研究问题
- RQ1在重复第二价格拍卖中,买家如何在同时满足预算和ROI约束的前提下实现最优出价?
- RQ2在联合预算和ROI约束下,最优事后出价策略的结构是什么?
- RQ3基于学习的出价算法是否能在随机竞争下实现次线性遗憾,同时保持约束满足?
- RQ4当面对同时具有预算和ROI约束的买家时,卖家的收益函数行为如何?
- RQ5通过在分阶段二分查找框架中适应买家的学习动态,定价算法是否能实现次线性遗憾?
主要发现
- 买家的最优事后出价策略是在估值与支出之比超过特定阈值的所有拍卖中获胜。
- 所提出的买家出价框架在随机竞争下保证了预算和ROI约束的满足。
- 买家的算法相对于最优事后策略实现了次线性遗憾。
- 卖家的收益函数表现出钟形结构,从而支持高效优化。
- 卖家的定价算法通过使用分阶段二分查找,并在每个阶段允许买家适应,实现了次线性遗憾。
- 买家学习与卖家定价之间的相互作用实现了稳定且收益最优的价格发现,且遗憾为次线性。
更好的研究,从现在开始
从阅读论文到最终审阅,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。