Remote Config 个性化
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。
通过 Remote Config 个性化设置,您可以自动为每位用户选择 Remote Config 参数,以针对目标进行优化。对参数进行个性化设置就像执行自动化、个体化并持续优化的永久性 A/B 测试。
当您在应用中使用 Remote Config 个性化功能时,可以为每位用户自动提供几种备选用户体验(针对您选择的目标进行了优化的备选方案)中的一种,从而为他们提供更具吸引力的体验。您可以使用 Remote Config 定位条件,根据特定的用户群组来确定个性化的 Remote Config 参数。
您可以针对能够使用 Google Analytics 来衡量的任何目标进行优化,并根据事件数量或某个事件参数的汇总值(总和)完成优化过程。这包括以下内置指标:
- 用户互动时长(根据用户互动时长进行优化)
- 广告点击次数(根据广告点击事件的总数进行优化)
- 广告展示次数(根据广告展示次数进行优化)
或者,您可以根据任何 Analytics 事件来针对自定义指标进行优化。以下是一些可能的事件:
- Play 商店或 App Store 评分提交
- 用户成功完成特定任务(例如通过游戏关卡)
- 应用内购买事件
- 电子商务事件,例如将商品添加到购物车、开始或完成结账
- 应用内购买和广告收入
- 虚拟货币支出
- 链接和内容分享以及社交网络活动
如需详细了解更多可能的个性化用例,请参阅 Remote Config 个性化功能有哪些用途?
开始
工作原理
个性化运用机器学习技术来为每个用户确定出色体验。在学习不同类型用户的出色体验,以及利用该知识来最大限度提升您的目标指标之间,该算法可以高效地进行权衡。系统会自动将个性化结果与从所提供的备用方案中始终获得随机体验的对照用户组进行比较 - 这种比较可以展示个性化系统带来了多少“提升”(增量值)。
如需详细了解 Remote Config 个性化算法和概念,请参阅 Remote Config 个性化简介。
实现流程
- 实现两个或更多您预期对某些用户而言感受最佳但对其他用户则不然的备选用户体验。
- 使用 Remote Config 参数,实现远程配置这些备选用户体验。请参阅 Remote Config 使用入门和 Remote Config 加载策略。
- 为参数启用个性化设置。Remote Config 会为每个用户分配对他们而言感受较佳的体验。请参阅入门指南。
个性化与 A/B 测试
A/B 测试旨在找出单一性能出色的用户执行体验,个性化与之不同,它尝试通过为每个用户动态地选择出色的用户体验来实现目标最大化。对于许多类型的问题,个性化都可以生成最佳结果,但 A/B 测试仍然有其用武之地:
首选个性化 |
首选 A/B 测试 |
当每位用户都可以从个性化的用户体验中受益时 |
当您希望为所有用户或特定用户群体提供一种最佳体验时 |
当您希望持续优化个性化模型时 |
当您需要在固定的时间范围内进行测试时 |
当您的优化目标可以简单地表述为分析事件的权重总和时 |
当您的优化目标需要仔细评估多个不同的竞争指标时 |
当您希望针对某个目标进行优化而不考虑任何权衡时 |
当您希望确定是否有某个变体对另一个变体表现出具有统计意义的显著提升,然后再推广时 |
当不必或不需要手动审核结果时 |
当需要手动审核结果时 |
例如,假设您希望在提示用户对您的应用进行评分之后,最大限度提升在 Play 商店中执行此操作的用户的数量。一个可能影响成功的因素是发送提示的时机:您是在用户首次、第二次还是第三次打开您的应用时显示提示?或者,您是否在用户成功完成某些任务时向他们显示提示?理想的时机可能由于各个用户的不同而有所变化:某些用户可能已准备好为您的应用评分,而另一些用户可能还需要更多时间。
优化反馈提示的时机是个性化的理想使用情形:
- 对于每个用户而言,最佳设置可能会有所不同。
- 借助 Analytics,可以轻松衡量出是否成功。
- 相关的用户体验的变化存在的风险极低,您甚至可能不需要进行权衡,也不需要进行手动审核。
试用
开始
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):2025-08-14。
[null,null,["最后更新时间 (UTC):2025-08-14。"],[],[],null,["\u003cbr /\u003e\n\nWith Remote Config personalization, you can automatically select\nRemote Config parameters for each user to optimize for an objective.\nPersonalizing a parameter is like performing an automatic, individualized,\ncontinuously-improving, and perpetual A/B test.\n\nWhen you use Remote Config personalization in your apps, you create more\nengaging experiences for each of your users by automatically providing them with\none of several alternative user experiences---the alternative that optimizes\nfor the objective you choose. You can target your personalized Remote Config\nparameters to specific user groups using\n[Remote Config targeting conditions](/docs/remote-config/parameters#conditions_rules_and_conditional_values).\n\nYou can optimize for any objective that's measurable using\nGoogle Analytics, and optimize by number of events or by the aggregated\nvalue (sum) of an event parameter. This includes the following built-in metrics:\n\n- User engagement time, which optimizes by user engagement time\n- Ad clicks, which optimizes by total number of ad click events\n- Ad impressions, which optimizes by the number of ad impressions\n\nOr, you can optimize for custom metrics based on any Analytics event. Some\npossibilities include:\n\n- Play Store or App Store rating submissions\n- User success at particular tasks, like completing game levels\n- In-app purchase events\n- E-commerce events, like adding items to a cart, or beginning or completing checkout\n- In-app purchase and ad revenue\n- Virtual currency spend\n- Link and content sharing and social networking activity\n\nFor more information about potential personalization use cases, see\n[What can I do with Remote Config personalization?](/docs/remote-config/personalization/use-cases)\n\n[Get started](/docs/remote-config/personalization/get-started)\n\nHow does it work?\n\nPersonalization uses machine learning to determine the optimal experience for\neach of your users. The algorithm efficiently trades off between learning the\nbest experience for different types of users and making use of that knowledge to\nmaximize your objective metric. Personalization results are automatically\ncompared to a holdout group of users who receive a persistent random experience\ndrawn from your provided alternatives---this comparison shows how much\n\"lift\" (incremental value) is generated by the personalization system.\n\nFor more information about Remote Config personalization algorithm and concepts,\nsee\n[About Remote Config personalization](/docs/remote-config/personalization/about).\n\nImplementation path\n\n1. Implement two or more alternative user experiences that you expect will be optimal for some users but not others.\n2. Make these alternatives remotely configurable with a Remote Config parameter. See [Get started with Remote Config](/docs/remote-config/get-started) and [Remote Config loading strategies](/docs/remote-config/loading).\n3. Enable personalization for the parameter. Remote Config will assign each of your users the experience that's optimal for them. See the [Getting started](/docs/remote-config/personalization/get-started) guide.\n\nPersonalization vs. A/B testing\n\nUnlike A/B tests, which are designed to find a single best performing user\nexperience, personalization attempts to maximize an objective by dynamically\nchoosing an optimal user experience for each user. For many types of problems,\npersonalization produces the best results, but A/B testing still has its uses:\n\n| Personalization preferred | A/B testing preferred |\n|-------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------|\n| When each user could benefit from a personalized user experience | When you want a single optimum experience for all users or a defined subset of users |\n| When you want to continuously optimize the personalization model | When you want to conduct tests during a fixed time window |\n| When your optimization goal can be expressed simply as a weighted sum of analytics events | When your optimization goal requires thoughtful evaluation of several different competing metrics |\n| When you want to optimize for an objective regardless of any trade-offs | When you want to determine if one variant shows a statistically significant improvement over another before rolling it out |\n| When manual review of results is not required or desired | When manual review of results is desirable |\n\nFor example, suppose you want to maximize the number of users who rate your app\nin the Play Store when you prompt them to. One factor that might contribute to\nsuccess is the timing of your prompt: do you show it when the user opens your\napp for the first, second, or third time? Or do you prompt them when they\nsuccessfully complete certain tasks? The ideal timing likely depends on the\nindividual user: some users might be ready to rate your app right away, while\nothers might need more time.\n\nOptimizing the timing of your feedback prompt is an ideal use case for\npersonalization:\n\n- The optimal setting is likely different for each user.\n- Success is easily measurable using Analytics.\n- The UX change in question is low risk enough that you probably don't need to consider trade-offs or conduct a manual review.\n\nTry it\n\n[Get started](/docs/remote-config/personalization/get-started)"]]