大规模观众留存分析背后的故事
[
[
Making videos is hard! Analyzing viewer retention and engagement of your videos is even harder. At Vimeo, we handle well over a billion videos of varying lengths and diverse viewing patterns, constantly growing in scale and complexity. Providing viewer retention analytics and AI-driven insights at second-level granularity instantly for any timeframe (historical, monthly, daily) and video length is crucial.
制作视频很难！分析观众对你的视频的保留率和参与度更难。在Vimeo，我们处理着超过十亿个不同长度和多样观看模式的视频，规模和复杂性不断增长。提供观众保留分析和基于AI的洞察，能够在任何时间框架（历史、每月、每日）和视频长度上以秒级粒度即时获取，这是至关重要的。
In this post I outline Vimeo’s architecture and approach for delivering viewer retention analytics with millisecond-level latency, discuss how we leverage AI for actionable insights, and highlight key challenges faced at scale.
在这篇文章中，我概述了Vimeo的架构和提供毫秒级延迟的观众保留分析的方法，讨论了我们如何利用AI获得可操作的洞察，并强调了在规模化过程中面临的主要挑战。
What is viewer retention and why does it matter?
什么是观众留存率，为什么它很重要？
Think of viewer retention as the measurement of a video’s ability to hold attention. This metric reveals the extent to which the audience engages with your video, helping to understand exactly which moments keep them hooked, or even cause them to leave.
将观众留存率视为视频保持注意力的能力的衡量标准。该指标揭示了观众与您视频的互动程度，帮助了解哪些时刻让他们保持关注，甚至导致他们离开。
At Vimeo, we visualize viewer retention as a simple XY graph (see Figure 1). The x-axis represents the video’s progression in seconds, while the y-axis indicates the number of views per second. You can even break down viewer retention data by specific users, dimensions (such as region, device domain), or particular time periods.
在Vimeo，我们将观众留存可视化为一个简单的XY图（见图1）。x轴表示视频的进度（以秒为单位），而y轴表示每秒的观看次数。您甚至可以按特定用户、维度（如地区、设备域）或特定时间段细分观众留存数据。
Figure 1. Viewer retention sample with retention going down as the video progresses. Note the AI insights box in the upper right.
图1. 观众留存样本，随着视频的进行...