2022年8月2日
By Damien读高级副总裁-数据产品
博客

流媒体战争的新时代已经到来，这场战争正变得个人化

在激烈竞争中, 流媒体服务正在寻求新的个性化策略来增加内容的参与度, 解决生产, 增加广告收入, 并鼓励用户返回他们的应用程序. What new and emerging data-driven approaches should streaming services use to boost their content discovery user experience?

流媒体服务正感受到竞争加剧的压力，这已经不是什么秘密了, 以及全球家庭预算的通胀压力. 流媒体之战已经演变成一场持续不断的争夺客户留存率的激战. 当谈到让用户相信流媒体服务的价值时, 内容参与是新的王者.

用户流失一直是订阅服务面临的挑战，但当就连Netflix也开始转向广告赞助的选择，你知道现有的减少流失的方法是不够的. Advertising-based business models bring their own customer retention and content discovery challenges, 当然, just ask the services pushing at the vanguard of the current trend for free ad-supported streaming TV (FAST).

人际接触值得付出努力

那么答案是什么呢? 添加越来越多的内容是件好事，但随后消费者会发现自己的选择太多了. 帮助他们尽快找到有吸引力的东西是至关重要的. Recommending content based on a user’s viewing history is a great way to shorten the time to content in the streaming user journey. 这是对已有数据的合理利用, 这对客户满意度有很大帮助, 不管是什么商业模式.

想要证明个性化真的可以改变局面? 来自我们自己客户的汇总数据表明了这一点, 平均, users who interact with sections of an app featuring personalization watch a greater duration of content, and engage with a larger variety of individual content items than those who only interact with generic sections. One of our customers saw the average number of content plays per user session rise by 133% when users were interacting with recommended content!

元数据加用户数据:推荐的艺术

流媒体服务面临的挑战是实现个性化并不是一件小事. 它既是一门艺术也是一门科学，需要元数据和行为洞察力的复杂结合. 元数据可以帮助你将一部犯罪剧与其他犯罪剧进行匹配, 或者一部汤姆·克鲁斯的电影和另一部他主演的电影. 另一方面, 行为洞察会告诉你，你的观众看过的唯一一部汤姆·克鲁斯的电影是 遥远的地方 从1992年开始，所以也许他们对历史浪漫更感兴趣而不是 使命:不可能的. Or that a significant proportion of people who watch your new gardening show are also watching a particular cooking show, 让它们成为交叉推广的理想选择.

个性化有什么新变化?

24i is seeing a surge of interest from both subscription and ad-funded services in new and emerging ways to make user-specific adjustments to their streaming user experience. Here are five questions to ask yourself to determine if your streaming service is making the most of modern personalization methodologies:

您是否使用数据来确定编辑派生内容的顺序?

算法推荐常常被编辑团队视为一种威胁. 我过去在天空的流媒体服务NOW TV(今天简称为NOW TV)工作 N OW), we’d regularly see the editorial and data teams sending out competing emails to see which would perform better. 这种“他们和我们”的态度是对天才的浪费. 如果你的编辑团队还没有充分利用数据科学的力量, 那么我的第一个也是最重要的建议就是尽可能快地开始.

It’s true that no algorithm can know your content (or your specific business goals) as well as your editorial team. 当然，同样地，没有编辑团队能够单独了解每个消费者. 在一起，你会得到两全其美. 从你的英雄旗帜开始. 让编辑团队选择您想要展示的内容, then use personalization algorithms to determine the order in which to display that content for each user based on their viewing history and what is most likely to appeal to them. 这是一个快速的胜利，你会惊讶于它对你的参与产生的即时影响.

您是否考虑过在标记数据驱动内容部分的方式背后的心理学?

许多流媒体服务都有“为你推荐”或“趋势”轨道. 数据驱动部分的其他常见标题包括“最受欢迎”,“更像这样。,”“因为你在看,”等. 这些方法都可能取得成功，但有些方法在应用的某个部分会比另一个部分更有效. 虽然有些标题看起来可以互换相同的内容, 当它们与细微差别的数据集配对时，效果最好.

如果你对个性化是认真的, you need to think carefully about the psychological impact of your section names and make sure they are suited to both your specific user base and your goals. Expert help and a lot of A/B testing will help you find the optimum mix for your service and your goals. 例如, our experience suggests you should use different algorithmic models for “Most Popular” and “Trending.“同样, a “Recommended For You” message is most effective when you want to help users discover something new. 在商业术语中, 这意味着鼓励用户从库中观看更广泛的内容.

你是否在利用“游戏后时刻”来鼓励更多的观看?

韦氏词典指出，“binge-watching”这个词最早出现在2003年. In less than 20 years, it’s become a common concept for a large proportion of the global population. 自动播放系列中的下一集是保持消费者使用你的服务的好方法, 这就是为什么它被各种尺寸的彩带广泛使用的原因. 然而，在这个系列的最后会发生什么呢? 或者在一部没有明显续集的一次性电影或纪录片的结尾?

Too many streaming services have gone all-in for binge-watching but are still wasting a prime opportunity to make personalized content recommendations at the same post-play moment for their non-episodic content. 如果你已经有了“因为你看了”的算法, 这应该是一个非常快速的胜利实施.

您的元数据是否经过优化以确保最佳的内容到内容推荐?

元数据是, 坦率地说, 在推荐方面，大多数流媒体服务面临的最大障碍. Every single company I speak to tells me they’d like to improve the quality of their metadata because that’s how you make good content-to-content matches. 在这个领域，质量比数量更重要. 有一家公司自豪地告诉我们，他们有超过50个,000个不同的关键词, 但是，如果每个关键字只用于库中的一两个内容, 这如何帮助你找到相关的匹配呢?

没有什么灵丹妙药可以提高元数据的质量, but we have made significant improvements for customers by using advanced metadata-prediction models (machine learning). 第一步是分析他们现有的元数据, 然后我们定义一个结构化的关键字框架——不要太多! 然后，机器可以填补空白，整理整个图书馆的关键字使用情况.

Have you taken personalization beyond the app and into your marketing messages to drive repeat visits?

Encouraging existing users to remember your app via marketing messages is an important step whether you’re looking to combat churn or increase ad-revenue. Integrating personalized content recommendations into those same messages takes you a step further; it gives the user a concrete reason to return. 如果你没有在邮件和推送通知中使用个性化的内容推荐, 那么现在是时候考虑一下了.

有很多变量需要考虑, 尤其是在你们交流的时间上, 但如果你做对了, 你可以看到很大的影响. One of our customers conducted A/B testing that compared the success rate of a regular email newsletter versus an email containing personalized content recommendations. 收到个性化邮件的那一组的转化率高出了121%，令人瞠目结舌.

个性化是一个永无止境的故事

The specific blend of data models and algorithms that are ideal for your streaming service can only be determined through regular testing and refinement. 这就是为什么推荐不能是一个简单的“设置它然后忘记它”的过程. They need continuous tuning—not least because your content library and user base is likely to shift over time. That is also why we offer personalization as a “managed service” so we can work continually to make incremental gains every month and adapt as content and UX changes.

接下来的几个星期, 流媒体 will publish a series of further 文章 in which I’ll break down some of the strategies outlined above in more detail. 如果你等不了那么久，你可以看看我们的电子指南: 现在，每个流媒体服务都应该采用五种提高参与度的策略.

[编者注:这是来自 24i. 流媒体接受供应商署名完全基于它们对我们读者的价值.]