首页 > AI前沿 > Serving-Aligned Entire-Space Conversion Modeling via Click-Through Rate Knowledge Distillation

Serving-Aligned Entire-Space Conversion Modeling via Click-Through Rate Knowledge Distillation

arXiv机器学习 2026-11-11 15:25 19 阅读 查看原文
用户要求将一段纯文本新闻正文排版成语义化的HTML片段。我需要遵循以下规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 2. 每个自然段落用

包裹,长文用

/

划分小节。 3. 列表内容用
  • ,引用用
    ,代码用
    ,重要句子可用。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出//等文档级标签。
    
    正文内容是一段关于推荐系统的技术论文摘要,没有明显的列表、引用或代码,但有重要概念。
    
    分析正文:
    "Recommendation systems may use predicted click-through conversion rate (pCTCVR), the product of predicted click-through rate (pCTR) and post-click conversion rate (pCVR), to guide ranking decisions."
    "Many pCVR models build on the Entire Space Multi-Task Model (ESMM), which jointly trains pCVR and auxiliary pCTR towers using pCTR and pCTCVR losses, but deploys only the pCVR tower."
    "At serving time, this tower may be paired with a separately trained pCTR model, creating a training-serving mismatch in pCTCVR when its predictions differ from those of the auxiliary pCTR tower."
    "In this paper, we use knowledge distillation (KD) from logged serving pCTR scores to supervise ESMM's auxiliary pCTR tower and improve training-serving pCTCVR alignment."
    "We denote variants by KD-$α$, where $α$ controls the hard-label loss weight."
    "On the Ali-CCP dataset, KD-0.3, the best variant on serving pCTCVR, increases ROC-AUC not only for serving pCTCVR but also for pCVR by 2.1% and 1.6%, respectively, and PR-AUC by 9.9% and 8.8%, respectively, relative to the ESMM baseline."
    "We also test negative downsampling with 5% nonclick retention."
    "Downsampled KD-0.3 increases mean AUC for serving pCTCVR and pCVR by 1.2% and 0.9%, respectively, -and PR-AUC by 8.5% and 4.5%, respectively, relative to downsampled ESMM."
    
    这可以分成几个部分:
    1. 背景介绍:推荐系统使用pCTCVR,ESMM模型及其训练-服务不匹配问题。
    2. 方法:使用知识蒸馏(KD)来改进。
    3. 实验结果:在Ali-CCP数据集上的结果,包括负下采样。
    
    可以用

    或

    来划分小节,比如:

    背景与问题

    ...

    ...

    ...

    方法

    ...

    ...

    实验结果

    ...

    ...

    ...

    检查是否需要加,可以将重要术语加粗,比如pCTCVR、ESMM、knowledge distillation (KD)等,但要注意不得增删改内容,加粗只是格式,不算改内容。 注意原文中的 "-and" 应该是原文中的连字符,保留原文。原文是 "-and PR-AUC",可能是个笔误,但必须保留原文,不能修改。 排版:

    背景与挑战

    Recommendation systems may use predicted click-through conversion rate (pCTCVR), the product of predicted click-through rate (pCTR) and post-click conversion rate (pCVR), to guide ranking decisions.

    Many pCVR models build on the Entire Space Multi-Task Model (ESMM), which jointly trains pCVR and auxiliary pCTR towers using pCTR and pCTCVR losses, but deploys only the pCVR tower.

    At serving time, this tower may be paired with a separately trained pCTR model, creating a training-serving mismatch in pCTCVR when its predictions differ from those of the auxiliary pCTR tower.

    方法

    In this paper, we use knowledge distillation (KD) from logged serving pCTR scores to supervise ESMM's auxiliary pCTR tower and improve training-serving pCTCVR alignment.

    We denote variants by KD-$α$, where $α$ controls the hard-label loss weight.

    实验结果

    On the Ali-CCP dataset, KD-0.3, the best variant on serving pCTCVR, increases ROC-AUC not only for serving pCTCVR but also for pCVR by 2.1% and 1.6%, respectively, and PR-AUC by 9.9% and 8.8%, respectively, relative to the ESMM baseline.

    We also test negative downsampling with 5% nonclick retention.

    Downsampled KD-0.3 increases mean AUC for serving pCTCVR and pCVR by 1.2% and 0.9%, respectively, -and PR-AUC by 8.5% and 4.5%, respectively, relative to downsampled ESMM.

    检查规则: - 只输出HTML片段本身,不要Markdown围栏、不要解释。 ✅ - 每个自然段落用

    包裹,长文用

    /

    划分小节。 ✅ 使用了

    - 列表内容用
    • ,引用用
      ,代码用
      ,重要句子可用。 ✅ 使用了和(KD-$α$ 可以视为代码或变量,用合适,不过原文是数学公式,用或都行,我用包裹KD-$α$,其余$α$保持原文。注意原文中KD-$α$和$α$,不要改变。用KD-$α$ 可能不太合适,因为是数学公式,但规则里说代码用 
                           
      ,这里不是代码。那就不用,用或者直接文本。为了安全,用包裹变量名。或者直接原样输出。我决定用包裹关键术语,变量保持原样。)
      
      修改方法部分:
      

      We denote variants by KD-$α$, where $α$ controls the hard-label loss weight.

      检查原文是否有增删改: "Recommendation systems may use predicted click-through conversion rate (pCTCVR), the product of predicted click-through rate (pCTR) and post-click conversion rate (pCVR), to guide ranking decisions." -> 保留 "Many pCVR models build on the Entire Space Multi-Task Model (ESMM), which jointly trains pCVR and auxiliary pCTR towers using pCTR and pCTCVR losses, but deploys only the pCVR tower." -> 保留 "At serving time, this tower may be paired with a separately trained pCTR model, creating a training-serving mismatch in pCTCVR when its predictions differ from those of the auxiliary pCTR tower." -> 保留 "In this paper, we use knowledge distillation (KD) from logged serving pCTR scores to supervise ESMM's auxiliary pCTR tower and improve training-serving pCTCVR alignment." -> 保留 "We denote variants by KD-$α$, where $α$ controls the hard-label loss weight." -> 保留 "On the Ali-CCP dataset, KD-0.3, the best variant on serving pCTCVR, increases ROC-AUC not only for serving pCTCVR but also for pCVR by 2.1% and 1.6%, respectively, and PR-AUC by 9.9% and 8.8%, respectively, relative to the ESMM baseline." -> 保留 "We also test negative downsampling with 5% nonclick retention