加入新推出的
Discord 社区,展开实时讨论,获得同行支持,并直接与 Meridian 团队互动!
覆盖面和频次
使用集合让一切井井有条
根据您的偏好保存内容并对其进行分类。
使用覆盖面和频次是确保广告系列取得良好效果的关键因素,但由于某些传统媒体渠道缺乏准确的覆盖面和频次指标,目前的营销组合模型 (MMM) 往往不会考虑到这一点。通常,MMM 依靠展示次数作为输入,但忽略了如下事实:用户可能会多次看到广告,并且具体的影响可能会因曝光频率而异。为了克服这一局限性,Meridian 提供了相关选项,可根据覆盖面和频次数据(而非单个执行指标)对任何媒体渠道的效应进行建模分析。这种方法可能会更精确地估计营销对业务成果的影响,并有助于根据频次建议优化广告系列的投放。
为便于纳入模型,覆盖面和频次数据必须与 KPI 和控制数据具有相同的地理位置级别和时间粒度。
此外:
媒体效应是对预期结果的叠加贡献。对于具有覆盖面和频次数据的渠道, \(i^{th}\) 渠道在地理位置 \(g\) 和时间段 \(t\) 的媒体效应可按以下方式建模分析:
$$
\beta_{g,i}^{[RF]} \text{Adstock} \left(\left\{
r_{g,t-s,i}^{[RF]} \text{Hill} \left(
f_{g,t-s,i}^{[RF]};\ ec_i^{[RF]}, \text{slope}_i^{[RF]}
\right)
\right\}_{s=0} ^L;\ \alpha_i^{[RF]} \right)
$$
其中:
- \(f_{g,t,i}^{[RF]}\) 是平均频次
- \(r_{g,t,i}^{[RF]}=L_{g,i}^{[RF]}(\overset {\cdot \cdot}
r_{g,t,i})^{[RF]}\) 是转换后的覆盖面。这是按人口比例和渠道中位数值调整后的值。如需了解详情,请参阅输入数据。
计算这种效应时,应首先对平均频次 \(f_{g,t,i}^{[RF]}\) 应用 Hill 函数,以调整饱和效应。接下来用每个地理位置和周经过 Hill 转换的频次乘以转换后的覆盖面。然后用 Adstock 函数对这些值进行加权,以捕获媒体曝光随时间产生的滞后效应。
根据 Hill 函数,媒体效应是频次的 S 型函数,这意味着出于成本效益考虑,最佳的平均覆盖频次可能大于 1。S 型曲线反映了这样一种直觉:从每次展示所带来的增量效果值考虑,可能存在一个最佳频次。可能需要达到一定的最低频次,才能强化品牌回想;而过高的频次可能会导致用户对广告产生疲劳,造成回报递减。
假设在频次保持不变的情况下,覆盖面与 KPI 之间存在线性关系。这意味着,每多覆盖一位用户,对 KPI 的影响都与之前覆盖的用户相同。只要媒体渠道的覆盖范围不超出其预期目标受众群体,这一假设就是合理的估计,目标受众群体以外的用户可能不太会受到该渠道的影响。线性覆盖面假设还有助于避免出现模型过度形参化、形参不可识别和马尔可夫链蒙特卡洛 (MCMC) 收敛问题。请注意,不要将这种线性效应外推到远远超出数据中观测到的覆盖面值范围。
如需详细了解覆盖面和频次,请参阅纳入了覆盖面和频次数据的贝叶斯分层媒体组合模型。
出于投资回报率考虑假设的频次与出于优化考虑假设的频次之间的差异
出于投资回报率考虑假设的频次与出于优化考虑假设的频次之间存在差异。您可以根据需要调整出于优化考虑假设的频次。
如投资回报率、边际投资回报率和响应曲线中所述,投资回报率衡量的是渠道在 MMM 有数据的时间窗口内执行时的投资回报。渠道的执行方式包括如何在各个地理位置和时间段分配展示次数,还包括该渠道的历史频次。
优化时会假设未来的广告系列会以最佳频次投放,因为频次通常由广告客户掌控,尤其是在数字渠道中。如果最佳频次与历史频次不同,那么从投资回报率来看,采用优化后的预算分配比例时渠道的效果可能与渠道的历史效果不一致。如果当前频次与最佳频次相差甚远,这种情况可能会更加严重。
如果未来的广告系列不以最佳频次投放,您可以使用优化选项来更改假设的频次。这对于无法以特定平均频次执行的渠道很有帮助。
如未另行说明,那么本页面中的内容已根据知识共享署名 4.0 许可获得了许可,并且代码示例已根据 Apache 2.0 许可获得了许可。有关详情,请参阅 Google 开发者网站政策。Java 是 Oracle 和/或其关联公司的注册商标。
最后更新时间 (UTC):2025-08-22。
[null,null,["最后更新时间 (UTC):2025-08-22。"],[[["\u003cp\u003eMeridian allows modeling media channels based on reach and frequency data, potentially leading to more accurate marketing impact estimates and optimized campaign execution.\u003c/p\u003e\n"],["\u003cp\u003eReach and frequency data should have the same geo and time granularity as sales and control data, with reach representing unique individuals exposed and frequency calculated as total impressions divided by reach.\u003c/p\u003e\n"],["\u003cp\u003eThe media effect model considers adstock, reach, and frequency, with the Hill function adjusting for saturation effects and potential diminishing returns from excessive frequency.\u003c/p\u003e\n"],["\u003cp\u003eReach is assumed to have a linear relationship with sales response, although diminishing marginal returns can occur with larger reach; Meridian restricts reach effect to linear to avoid model complexity and potential issues.\u003c/p\u003e\n"],["\u003cp\u003eThere are differences in assumed frequency for ROI calculation and optimization, allowing adjustments for future campaigns to reflect potential changes in frequency execution.\u003c/p\u003e\n"]]],["Meridian enhances marketing mix models (MMMs) by incorporating reach and frequency data, unlike traditional MMMs that only use impressions. This approach models each channel's effect using unique individual reach and average frequency per time period and location. The model employs a Hill function to account for frequency saturation and an Adstock function for lagged effects. While frequency's impact is S-shaped, reach's impact is assumed linear. Optimization allows adjusting the assumed frequency to account for varying execution capabilities.\n"],null,["# Reach and frequency\n\nThe use of reach and frequency is a crucial factor in effective ad campaigns,\nbut it is not often considered in current marketing mixed models (MMMs) due to\nthe lack of accurate reach and frequency metrics for some traditional media\nchannels. Typically, MMMs rely on impressions as input, neglecting the fact that\nindividuals can be exposed to ads multiple times, and the impact can vary with\nexposure frequency. To overcome this limitation, Meridian offers the\noption to model any media channel's effect based on reach and frequency data,\ninstead of a single execution metric. This approach can potentially yield more\nprecise estimates of marketing impact on business outcomes and aid in optimizing\ncampaign execution through frequency recommendations.\n\nFor modeling purposes, the reach and frequency data must be at the same level of\ngeo and time granularity as the KPI and controls data.\n\nAdditionally:\n\n- The reach data should be the number of unique individuals exposed to the\n channels' ad within each time period instead of the cumulative number of\n individuals reached over consecutive time periods.\n\n- The frequency data should be the total number of impressions divided by the\n reach for each time period.\n\nThe *media effect* is the additive contribution to expected outcome. For channels\nwith reach and frequency data, the media effect of the \\\\(i\\^{th}\\\\) channel\nwithin geo \\\\(g\\\\) and time period \\\\(t\\\\) is modeled as follows: \n$$ \\\\beta_{g,i}\\^{\\[RF\\]} \\\\text{Adstock} \\\\left(\\\\left\\\\{ r_{g,t-s,i}\\^{\\[RF\\]} \\\\text{Hill} \\\\left( f_{g,t-s,i}\\^{\\[RF\\]};\\\\ ec_i\\^{\\[RF\\]}, \\\\text{slope}_i\\^{\\[RF\\]} \\\\right) \\\\right\\\\}_{s=0} \\^L;\\\\ \\\\alpha_i\\^{\\[RF\\]} \\\\right) $$\n\nWhere:\n\n- \\\\(f_{g,t,i}\\^{\\[RF\\]}\\\\) is the average frequency\n- \\\\(r_{g,t,i}\\^{\\[RF\\]}=L_{g,i}\\^{\\[RF\\]}(\\\\overset {\\\\cdot \\\\cdot} r_{g,t,i})\\^{\\[RF\\]}\\\\) is the transformed reach. This is scaled by population and the median value for the channel. For more information, see [Input\n data](/meridian/docs/basics/input-data).\n\nThis effect is calculated by first applying the Hill function to the average\nfrequency \\\\(f_{g,t,i}\\^{\\[RF\\]}\\\\) to adjust for saturation effects. The\nHill-transformed frequency for each geo and week is multiplied by transformed\nreach. These values are then weighted by the Adstock function to capture lagged\neffects of media exposure over time.\n\nThe Hill function allows for the media effect to be *S* shaped as a function of\nfrequency, which means that the optimal average reach for cost effectiveness may\nbe greater than one. The *S* shaped curve reflects the intuition that there\nmight be an optimal frequency for incremental outcome value per impression. A\ncertain minimum frequency might be necessary to reinforce brand recall, while\nexcessive frequency can result in ad fatigue and diminishing returns.\n\nReach is assumed to have a linear relationship with the KPI, while holding\nfrequency fixed. This means that each additional individual reached has the\nsame effect on the KPI as those reached previously. This assumption is a\nreasonable approximation as long as the media channel does not reach individuals\nwell beyond its intended target audience, who may be less affected by the\nchannel. The linear reach assumption also helps avoid model\noverparameterization, parameter non-identifiability, and Markov Chain Monte\nCarlo (MCMC) convergence issues. Be careful about extrapolating this linear\neffect far outside the range of reach values observed in the data.\n\nFor more information about reach and frequency, see [Bayesian Hierarchical Media\nMix Model Incorporating Reach and Frequency\nData](https://research.google/pubs/bayesian-hierarchical-media-mix-model-incorporating-reach-and-frequency-data/).\n\nDifferences between assumed frequency for ROI and for optimization\n------------------------------------------------------------------\n\nThere are differences between the assumed frequency for ROI and for\noptimization. You can adjust the assumed frequency for optimization if needed.\n\nAs discussed in [ROI, mROI, and response\ncurves](/meridian/docs/basics/roi-mroi-response-curves), ROI measures the return\nof investment of a channel as it was executed during the time window for which\nthe MMM has data. How a channel was executed includes how impressions are\nallocated across geos and time, and also includes the historical frequency of\nthat channel.\n\nOptimization assumes that future campaigns will be executed at the optimal\nfrequency, since frequency is something often in an advertiser's control,\nespecially for digital channels. If the optimal frequency is different from the\nhistorical frequency, a channel's performance in optimized budget allocation\nmight not match the channel's historical performance according to ROI. This can\nbe exacerbated if the current frequency is far from the optimal frequency.\n\nIf future campaigns won't be executed at the optimal frequency, you can use the\noptimization option to change the assumed frequency. This can be helpful for\nchannels that cannot be executed at a specific average frequency."]]