从数据到洞察：细分Airbnb的供应

Introduction

介绍

At Airbnb, our supply comes from hosts who decide to list their spaces on our platform. Unlike traditional hotels, these spaces are not all interchangeable units in a building that are available to book year-round. Our hosts are people, with different earnings objectives and schedule constraints — leading to different levels of availability to host. Understanding these differences is a key input into how we develop our products, campaigns, and operations.

在Airbnb，我们的供应来自决定在我们平台上列出其空间的主机。与传统酒店不同，这些空间并不是全年可预订的建筑中的可互换单元。我们的主机是人，有不同的收入目标和时间安排限制——导致不同的托管可用性水平。了解这些差异是我们开发产品、活动和运营的关键输入。

Over the years, we’ve created various ways to measure host availability, developing “features” that capture different aspects of how and when listings are available. However, these features provide an incomplete picture when viewed in isolation. For example, a ~30% availability rate could indicate two very different scenarios: a host who only accepts bookings on weekends, or a host whose listing is only available during a specific season, such as summer.

多年来，我们创建了各种方法来衡量房东的可用性，开发了捕捉房源何时和如何可用的“特征”。然而，当单独查看这些特征时，它们提供的图景是不完整的。例如，约 30% 的可用率可能表示两种截然不同的情况：一个只在周末接受预订的房东，或一个房源只在特定季节（如夏季）可用的房东。

This is where segmentation comes in.

这就是细分的作用所在。

By combining multiple features, segmentation allows us to create discrete categories that represent the different availability patterns of hosts.
通过结合多个特征，分割使我们能够创建离散的类别，代表不同的房东可用性模式。

But traditional segmentation methodologies, such as “RFM” (Recency, Frequency, Monetary), are focused on customer value rather than calendar dynamics, and are often limited to one-off analyses on small datasets. In contrast, we need an approach that can handle calendar data and daily inference for millions of listings.

但传统的细分方法，如“RFM”（最近、频率、货币），侧重于客户价值而非日历动态，且通常仅限于对小数据集的一次性分析。相比之下，我们需要一种能够处理日历数据和每日推断数百万列表的方法。

To address the above challenges, this blog post explores how Airbnb used segmentation to better understand host behavior at scale. By enrich...