← Back to System Design Mind Map
🏗️ System DesignID: esmm-sample-selection-bias

ESMM Entire Space Multi-Task Model

ESMM 全空间多任务模型
🎯Core Definition
ESMM (Entire Space Multi-Task Model, proposed by Alibaba Alimama at SIGIR 2018) is a multi-task learning architecture specifically engineered to resolve the twin industry dilemmas in Conversion Rate (CVR) estimation: Sample Selection Bias (SSB) and Data Sparsity (DS); using probability chain rules: pCTCVR=pCTR×pCVRp\text{CTCVR} = p\text{CTR} \times p\text{CVR}, ESMM trains CTR and CTCVR simultaneously across the entire un-gated impression space, implicitly deriving pCVR=pCTCVRpCTRp\text{CVR} = \frac{p\text{CTCVR}}{p\text{CTR}} without conditioning training exclusively on clicked samples.
💡Use Cases
E-commerce purchase conversion rate estimation, performance advertising bidding (pCVR), and multi-step conversion funnel modeling.
⚡Key Problems Solved
Conventional CVR models train solely on clicked items but serve across the full impression pool, triggering severe distribution shifts (SSB) and over-fitting on sparse positive data; ESMM's full-space joint training completely eliminates selection bias and data sparsity.
🎯Core Interview Exam Points
Fundamentals · Production Scenarios · Deep Dive
Core Def
Explain the core definition, underlying mechanism/math, and key principles of ESMM Entire Space Multi-Task Model.
Use Cases
In production engineering, what are the primary use cases for ESMM Entire Space Multi-Task Model and what core pain points does it solve?
🔥 High-Frequency Deep-Dive Questions
1
Derive the joint loss function of ESMM over full impression space combining CTR and CTCVR targets?
2
Why does online serving extract the raw output of the CVR Tower directly as the predicted pCVRp\text{CVR} without dividing by CTR?
3
How does embedding weight sharing between CTR and CVR towers leverage abundant CTR impressions to alleviate CVR data sparsity?
4
How does ESM2 extend the ESMM framework across multi-stage conversion DAG funnels (Impression -> Click -> Cart -> Purchase)?
5
Why does optimizing CVR via explicit division cause numerical instability, and how does multiplicative CTCVR loss avoid gradient explosions?
Updated 2026-08-14
🎯
Verify Mastery: Practice Exam Questions for "ESMM Entire Space Multi-Task Model"
Take pitfall multiple-choice questions with derivation options; mistakes are automatically logged.
🚀 Start Card Practice ➔
← Previous CardPLE Progressive Layered ExtractionNext Card →Feature Store Online/Offline Consistency

🔗 More System Design Knowledge Cards

引用出处: TalentMe 450+ 工业级 AI 架构考点知识库 (Tech Vault) · 遵循 CC-BY-4.0 知识共享协议
URL: https://talentme.airsota.com/resources/tech/roadmap/mindmap/System_Design/card/esmm-sample-selection-bias
中文速览与面试要点

ESMM 全空间多任务模型 (ESMM Entire Space Multi-Task Model)

定义: ESMM(Entire Space Multi-Task Model,由阿里妈妈算法团队在 SIGIR 2018 提出)是一种专为解决推荐与广告系统中转化率 (CVR) 预估两大业界核心难题——样本选择偏差 (Sample Selection Bias, SSB) 与数据稀疏 (Data Sparsity)——而设计的全空间多任务网络架构;根据概率链式法则,点击且转化概率等于点击率乘点击后转化率:p CTCVR = p CTR p CVR;ESMM 在全量曝光样本空间(全部展示记录)上同时建模与训练主任务 p CTR 和辅助任务 p CTCVR,通过二者隐式导出 p CVR = p CTCVRp CTR,彻底摆脱了传统 CVR 模型只能在“已点击样本子集”上训练的局限。 使用场景: 电商交易转化率预估、广告计费 pCVR 预估、O2O 下单转化率与多步转化漏斗建模。 解决的问题: 传统 CVR 模型仅用“点击样本”训练,但在线 Serving 却要在“全量曝光样本”上推理,造成极严重的样本分布偏移 (SSB);且由于点击率通常仅 2%~5%,点击后购买样本极其稀疏导致过拟合;ESMM 在全空间建模主任务与辅助任务,彻底攻克样本偏差与数据稀疏两大顽疾。

核心要点

  • 推导 ESMM 的联合损失函数公式 L( _ctr, _cvr) = _i=1^N l(y_i, f(x_i; _ctr)) + _i=1^N l(y_i \& z_i, f(x_i; _ctr) f(x_i; _cvr))?
  • 为什么在在线推理 (Serving) 阶段,系统只需要直接提取 CVR Tower 的输出值作为预估转化率 p CVR?
  • ESMM 结构中 CTR 塔与 CVR 塔底层共享 Embedding 查找表是如何极大缓解 CVR 塔样本稀疏问题的?
  • 针对多步转化漏斗(如 曝光 -> 点击 -> 加购 -> 购买),如何将 ESMM 进一步拓展为 ESM2 (有向无环图全空间多步转化网络)?
  • 分析为什么直接在训练中让 CVR 塔以除法形式优化可能会导致数值不稳定性(如 p CTR 0 时梯度爆炸)?