TL;DR,

摘要

Taobao, as the largest online retail platform in the world, provides billions of online display advertising impressions for millions of advertisers every day. For commercial purposes, the advertisers bid for specific spots and target crowds to compete for business traf- fic. The platform chooses the most suitable ads to display in tens of milliseconds. Common pricing methods include cost per mille (CPM) and cost per click (CPC). Traditional advertising systems target certain traits of users and ad placements with fixed bids, essentially regarded as coarse-grained matching of bid and traffic quality. However, the fixed bids set by the advertisers competing for different quality requests cannot fully optimize the advertisers’ key requirements. Moreover, the platform has to be responsible for the business revenue and user experience. Thus, we proposed a bid optimizing strategy called optimized cost per click (OCPC) which automatically adjusts the bid to achieve finer matching of bid and traffic quality of page view (PV) request granularity. Our approach optimizes advertisers’ demands, platform business rev- enue and user experience and as a whole improves traffic alloca- tion efficiency. We have validated our approach in Taobao display advertising system in production. The online A/B test shows our algorithm yields substantially better results than previous fixed bid manner.

问题背景

1. 淘宝同时作为需求方和供给方，能够拥有最全面的用户数据
2. 大部分广告主都是中小商家，他们更加关注受益的增加，而不是做品牌推广
3. 不同的广告主往往有不同的目标，曝光、点击、转化或者ROI，商家采用点击付费的方式进行竞价
4. 广告会有平台指标的要求，比如CTR、CVR或者GMV，我们以GMV为主进行分析。这样做，首先是因为我们希望商业化的流量并不会影响用户的体验，设置GMV目标对于平台和广告主是双赢的局面，其次平台会对商家采用近似固定比例抽成，提高GMV长远来看也有利于平台

解决方案

，商家基础出价$b_a$，笔单价$v_a$，单条流量的ROI可以写作 $\text{roi}_{u,a}=\frac{p(c|u,a)\cdot v_a}{b_a}$ 其中$p(c|u,a)$表示用户$u$点击广告$a$之后的转化率，商品的转化率为 $\text{roi}_a =\frac{v_a \cdot \sum_u n_u \cdot p(c|u,a)}{b_a\cdot \sum_u n_u} =\frac{\mathbb{E}_u[c|u,a] \cdot v_a}{b_a}$ 我们希望我们优化后的单次出价$b_a^*$能够保证ROI不降低，即 \begin{aligned} \text{roi}_{a} & \leq \text{roi}_a^* \\ \frac{\text{roi}_a }{\text{roi}_{a}^* } &\leq 1 \\ \frac{\frac{\mathbb{E}_u[c|u,a] \cdot v_a}{b_a}}{\frac{p(c|u,a)\cdot v_a}{b_a}^*} &\leq 1 \\ \frac{b_a^*}{b_a} &\leq \frac{p(c|u,a)}{\mathbb{E}_u[c|u,a]} \end{aligned} 为了保险起见，对出价范围进行限制，上下变动比例系数$r_a$，上下界为 \begin{aligned} &l(b_a^*) = \begin{cases} b_a * (1-r_a) && \frac{p(c|u,a)}{\mathbb{E}_u[c|u,a]} < 1 \\ b_a && \frac{p(c|u,a)}{\mathbb{E}_u[c|u,a]} \geq 1 \\ \end{cases}\\ &u(b_a^*) = \begin{cases} b_a && \frac{p(c|u,a)}{\mathbb{E}_u[c|u,a]} < 1 \\ b_a * \min\Big(1+r_a, \frac{p(c|u,a)}{\mathbb{E}_u[c|u,a]}\Big)&& \frac{p(c|u,a)}{\mathbb{E}_u[c|u,a]} \geq 1 \\ \end{cases} \end{aligned} 取值范围如下图所示

1. 将全部广告按照上界出价的平台收益$f(u(b_i^*))$排序
2. 选择当前排序中下界最高的广告，eCPM参竞价格为$t$
3. 找到排序中第一个上界大于该价格的，$u(s_k)\geq t$，选择这个商品
4. 然后排序中删除这个商品，继续2和3