🏗️ System DesignID: esmm-sample-selection-bias

ESMM Entire Space Multi-Task Model

Q: Derive the joint loss function of ESMM over full impression space combining CTR and CTCVR targets?

【ESMM Entire Space Multi-Task Model 核心考点 1】Derive the joint loss function of ESMM over full impression space combining CTR and CTCVR targets?。技术解析：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);... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/System_Design/card/esmm-sample-selection-bias

Q: Why does online serving extract the raw output of the CVR Tower directly as the predicted $p\text{CVR}$ without dividing by CTR?

【ESMM Entire Space Multi-Task Model 核心考点 2】Why does online serving extract the raw output of the CVR Tower directly as the predicted $p\text{CVR}$ without dividing by CTR?。技术解析：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);... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/System_Design/card/esmm-sample-selection-bias

Q: How does embedding weight sharing between CTR and CVR towers leverage abundant CTR impressions to alleviate CVR data sparsity?

【ESMM Entire Space Multi-Task Model 核心考点 3】How does embedding weight sharing between CTR and CVR towers leverage abundant CTR impressions to alleviate CVR data sparsity?。技术解析：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);... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/System_Design/card/esmm-sample-selection-bias

Q: How does ESM2 extend the ESMM framework across multi-stage conversion DAG funnels (Impression -> Click -> Cart -> Purchase)?

【ESMM Entire Space Multi-Task Model 核心考点 4】How does ESM2 extend the ESMM framework across multi-stage conversion DAG funnels (Impression -> Click -> Cart -> Purchase)?。技术解析：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);... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/System_Design/card/esmm-sample-selection-bias

Q: Why does optimizing CVR via explicit division cause numerical instability, and how does multiplicative CTCVR loss avoid gradient explosions?

【ESMM Entire Space Multi-Task Model 核心考点 5】Why does optimizing CVR via explicit division cause numerical instability, and how does multiplicative CTCVR loss avoid gradient explosions?。技术解析：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);... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/System_Design/card/esmm-sample-selection-bias

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:

p\text{CTCVR} = p\text{CTR} \times p\text{CVR}

, ESMM trains CTR and CTCVR simultaneously across the entire un-gated impression space, implicitly deriving

p\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.

🎯5 High-Frequency Exam Points

Derive the joint loss function of ESMM over full impression space combining CTR and CTCVR targets?

Why does online serving extract the raw output of the CVR Tower directly as the predicted

p\text{CVR}

without dividing by CTR?

How does embedding weight sharing between CTR and CVR towers leverage abundant CTR impressions to alleviate CVR data sparsity?

How does ESM2 extend the ESMM framework across multi-stage conversion DAG funnels (Impression -> Click -> Cart -> Purchase)?

Why does optimizing CVR via explicit division cause numerical instability, and how does multiplicative CTCVR loss avoid gradient explosions?

📖 In-depth Guide:📄 recommendation-system-design →

Updated 2026-08-14

🎯

Test Your Knowledge: Practice Questions for "ESMM Entire Space Multi-Task Model"

Single choice pitfall questions with instant feedback and mistake tracking.

🚀 Start Card Practice ➔

← Previous CardPLE Progressive Layered Extraction Next Card →Feature Store Online/Offline Consistency

🔗 More System Design Knowledge Cards

RecSys Multi-Stage Funnel & 50ms SLA DSSM Two-Tower Retrieval YouTube DNN Candidate Generation Pre-Ranking Lightweight Architecture