Pinterest的大规模用户序列

User Understanding team: Zefan Fu, Minzhe Zhou, Neng Gu, Leo Zhang, Kimmie Hua, Sufyan Suliman | Software Engineer, Yitong Zhou | Software Engineering Manager

用户理解团队：傅泽凡，周敏哲，顾能，张利奥，华金明，苏菲安-苏里曼|软件工程师，周奕彤|软件工程经理

Index Core Entity team: Dumitru Daniliuc, Jisong Liu, Kangnan Li | Software Engineer, Shunping Chiu | Software Engineering Manager

索引核心实体团队：Dumitru Daniliuc, Jisong Liu, Kangnan Li | 软件工程师, Shunping Chiu | 软件工程经理

Understanding and responding to user actions and preferences is critical to delivering a personalized, high quality user experience. In this blog post, we’ll discuss how multiple teams joined together to build a new large-scale, highly-flexible, and cost-efficient user signal platform service, which indexes the relevant user events in near real-time, constructs them into user sequences, and makes it super easy to use both for online service requests and for ML training & inferences.

了解并回应用户的行为和偏好，对于提供个性化的高质量用户体验至关重要。在这篇博文中，我们将讨论多个团队如何联合起来，建立一个新的大规模、高度灵活、低成本的用户信号平台服务，该服务以近乎实时的方式索引相关的用户事件，将其构建为用户序列，并使其超级容易用于在线服务请求和ML训练与推断。

Background & Context

背景和背景

User sequence is one type of ML feature composed as a time-ordered list of user engagement activities. The sequence captures one’s recent actions in real-time, reflecting their latest interests as well as their shift of focus. This kind of signal plays a critical role in various ML applications, especially for large-scale sequential modeling applications (see example).

用户序列是一种ML特征，由用户参与活动的时间顺序列表组成。该序列实时捕捉一个人最近的行动，反映了他们最新的兴趣以及他们的焦点转移。这种信号在各种ML应用中起着关键作用，特别是对于大规模的序列建模应用（见例子）。

To make the real-time user sequence more accessible within the Pinterest ML ecosystem, and to empower our daily metrics improvement, we list the following key features to deliver for ML applications:

为了使实时用户序列在Pinterest ML生态系统内更容易获得，并增强我们日常指标改进的能力，我们列出了以下为ML应用提供的关键功能：

• Real-time: on average < 2 seconds latency from a user’s latest action to the service response
• 实时性：从用户的最新行动到服务响应的平均延迟时间<2秒
• Flexibility: data can be fetched and reused by a mix-and-use pattern to enable faster iterations fo...