文章背景与核心概要
在大规模推荐系统中，内容陈旧是导致用户产生摩擦和投诉的主要原因之一。传统的应对方法（如粗暴的时间截断或滞后的互动启发式算法）无法准确捕捉相关性丢失的真实机制。本文介绍了 SDF（Supersession-Decay Filtering，替代-衰减过滤），这是一个在 Google Discover 工业级规模上部署的生产级双滤镜框架。
该框架在排序阶段的上游同时处理了替代（新更新使旧内容过时）和relevance decay（相关性衰减，即物品的信息价值在其生命周期中自然消退），从而成功提升了用户互动并降低了系统服务成本。经过两年的生产环境部署，用户提交的陈旧度投诉减少了 54.9%，证明了该范式在工业级推荐系统中的强大鲁棒性与可扩展性。
解构推荐系统中的内容陈旧度：用于替代与衰减的双滤镜框架 (Decomposing Staleness in Recommender Systems: A Dual-Filter Framework for Supersession and Decay)
执行摘要 (Executive Summary)
Content staleness is a primary source of user friction and complaints in large-scale recommender systems. Traditional countermeasures—such as crude age cutoffs or lagging engagement heuristics—fail to accurately capture the true mechanisms of relevance loss.
内容陈旧是大规模推荐系统中用户产生摩擦和投诉的主要根源。传统的应对措施（如粗暴的时间硬编码截断或滞后的互动启发式算法）无法准确捕捉相关性丢失的真实机制。
This paper introduces SDF (Supersession-Decay Filtering), a production-grade dual-filter framework deployed at industrial scale in Google Discover. By addressing both supersession (where newer updates render older content obsolete) and relevance decay (where an item's informational value naturally wanes over its lifecycle) upstream of the ranking stage, SDF successfully improves user engagement and reduces system serving costs. Over a two-year production deployment, user-submitted staleness reports declined by 54.9%.
本文介绍了 SDF（替代-衰减过滤，Supersession-Decay Filtering），这是一个在 Google Discover 中以工业级规模部署的生产级双滤镜框架。通过在排序阶段的上游同时解决替代效应（更新的内容使旧内容过时）和相关性衰减（物品的信息价值在其生命周期中自然消退）问题，SDF 成功提升了用户互动并降低了系统服务成本。在长达两年的生产环境部署中，用户提交的陈旧度投诉下降了 54.9%。
论文元数据 (Paper Metadata)
- arXiv ID: arXiv:2608.15780 [cs.IR]
- Related DOI: 10.1145/3799682.3840082
- Conference Track: CIKM Applied Research Track 2026
- Primary Subject: Information Retrieval (
cs.IR) - Secondary Subject: Artificial Intelligence (
cs.AI) - Submission Date: August 16, 2026
- Authors: Di Bai, Feng Han, Zhenwei Tang, Jintao Liu, Luoshu Wang, Jialu Liu
- arXiv ID: arXiv:2608.15780 [cs.IR]
- 相关 DOI: 10.1145/3799682.3840082
- 会议赛道: CIKM Applied Research Track 2026
- 主分类: 信息检索 (
cs.IR)- 次分类: 人工智能 (
cs.AI)- 提交日期: 2026年8月16日
- 作者: Di Bai, Feng Han, Zhenwei Tang, Jintao Liu, Luoshu Wang, Jialu Liu
摘要 (Abstract)
Stale recommendations are a pervasive challenge and a leading source of user complaints on large-scale content platforms. Items lose relevance through two primary mechanisms: supersession, where emerging updates render prior coverage stale, and relevance decay, where an item's informational value naturally diminishes over its lifecycle. Traditional countermeasures serve as crude proxies: age cutoffs poorly reflect actual relevance loss, while engagement heuristics rely on lagging signals, broadly exposing users to stale content before the system adapts.
陈旧的推荐内容是大规模内容平台上普遍存在的挑战，也是用户投诉的主要来源。物品通过两种主要机制失去相关性：一是替代效应（新兴的更新使先前的报道过时），二是相关性衰减（物品的信息价值在其生命周期中自然递减）。传统的应对措施只充当粗糙的替代指标：时间截断无法很好地反映实际的相关性损失，而互动启发式算法则依赖于滞后信号，在系统适应之前，广泛地将陈旧内容暴露给用户。
We present SDF (Supersession-Decay Filtering), a staleness filtering system fully deployed in Google Discover, a personalized recommendation feed with hundreds of millions of daily and billions of monthly active users. SDF targets both mechanisms with complementary filters, each powered by a learned model: 1. Relational Staleness Model: Detects supersession between item pairs. 2. Predicted Traffic Ratio (PTR) Model: Forecasts relevance decay from an item's content, trained on lifetime visit traffic.
我们提出了 SDF（Supersession-Decay Filtering），这是一个完全部署在 Google Discover 中的陈旧度过滤系统。Google Discover 是一个拥有数亿日活用户和数十亿月活用户的个性化推荐信息流。SDF 通过互补的滤镜针对这两种机制，每个滤镜都由一个学习模型驱动： 1. 关系型陈旧度模型（Relational Staleness Model）： 检测物品对之间的替代关系。 2. 预测流量比（PTR）模型： 根据物品的内容预测相关性衰减，该模型基于全生命周期的访问流量进行训练。
Applied via disjunction upstream of the ranking stage, SDF prunes stale candidates, measurably reducing downstream serving costs. Online experiments demonstrate that these filters significantly reduce the prevalence of stale content while improving user engagement. Over a two-year production deployment, user-filed staleness reports (in-product user feedback) declined by 54.9% relative to the pre-deployment baseline, establishing SDF as a robust and scalable paradigm for resolving content staleness at an industrial scale.
SDF 通过在排序阶段上游采用析取（disjunction）方式应用，能够剔除陈旧的候选内容，从而显著降低下游的服务成本。在线实验表明，这些滤镜显著降低了陈旧内容的流行度，同时提升了用户的互动率。在长达两年的生产部署中，相对于部署前的基线，用户提交的陈旧度投诉（产品内用户反馈）下降了 54.9%，这确立了 SDF 作为在工业级规模上解决内容陈旧问题的一个稳健且可扩展的范式。
访问与资源 (Access & Resources)
- Full-Text Options: View PDF | HTML Version | TeX Source
- Additional Tools: NASA ADS | Google Scholar | Semantic Scholar
