Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
97 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Entire Space Multi-Task Model: An Effective Approach for Estimating Post-Click Conversion Rate (1804.07931v2)

Published 21 Apr 2018 in stat.ML, cs.IR, and cs.LG

Abstract: Estimating post-click conversion rate (CVR) accurately is crucial for ranking systems in industrial applications such as recommendation and advertising. Conventional CVR modeling applies popular deep learning methods and achieves state-of-the-art performance. However it encounters several task-specific problems in practice, making CVR modeling challenging. For example, conventional CVR models are trained with samples of clicked impressions while utilized to make inference on the entire space with samples of all impressions. This causes a sample selection bias problem. Besides, there exists an extreme data sparsity problem, making the model fitting rather difficult. In this paper, we model CVR in a brand-new perspective by making good use of sequential pattern of user actions, i.e., impression -> click -> conversion. The proposed Entire Space Multi-task Model (ESMM) can eliminate the two problems simultaneously by i) modeling CVR directly over the entire space, ii) employing a feature representation transfer learning strategy. Experiments on dataset gathered from Taobao's recommender system demonstrate that ESMM significantly outperforms competitive methods. We also release a sampling version of this dataset to enable future research. To the best of our knowledge, this is the first public dataset which contains samples with sequential dependence of click and conversion labels for CVR modeling.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (7)
  1. Xiao Ma (169 papers)
  2. Liqin Zhao (17 papers)
  3. Guan Huang (75 papers)
  4. Zhi Wang (261 papers)
  5. Zelin Hu (3 papers)
  6. Xiaoqiang Zhu (27 papers)
  7. Kun Gai (125 papers)
Citations (373)

Summary

  • The paper presents a novel ESMM that leverages the entire impression space to accurately estimate post-click conversion rates.
  • It employs a multi-task framework to jointly model CTR, CVR, and CTCVR, achieving significant AUC improvements over traditional methods.
  • The study also releases a public dataset capturing sequential click-to-conversion events, encouraging further research in recommendation systems.

Analysis of the Entire Space Multi-Task Model for Post-Click Conversion Rate Estimation

The paper, "Entire Space Multi-Task Model: An Effective Approach for Estimating Post-Click Conversion Rate," presents a significant paper on the challenges and methodologies involved in predicting Post-Click Conversion Rate (CVR) in online recommender systems. The proposed model, termed the Entire Space Multi-task Model (ESMM), offers an innovative solution to simultaneously address the inherent problems of sample selection bias and data sparsity in CVR estimation.

The task of CVR estimation follows a user's interaction sequence — from impression to click and finally to conversion. Traditional approaches generally experience difficulties due to their reliance on models trained on biased, clicked samples, leading to inaccuracies when applied to the entire impression space. Furthermore, CVR data tends to be sparse because conversions are rare events relative to impressions and clicks.

ESMM's core contribution is leveraging the entire impression sample space during training rather than solely focusing on clicked samples. By employing a multi-task learning framework, ESMM simultaneously models the click-through rate (CTR), CVR, and combined CTR-CVR (CTCVR) probabilities, deriving CVR from the product of estimated CTR and CTCVR. This novel approach allows ESMM to effectively eliminate sample selection bias by training on all available impressions and employing shared parameter learning to mitigate data sparsity issues.

The paper's numerical results bolster the claim of ESMM's improved efficacy. In tests conducted on datasets from Alibaba's Taobao recommender system, ESMM consistently exhibited superior performance over conventional CVR models, with notable AUC improvements in both CVR prediction on clicked samples and CTCVR prediction across all samples. This comprehensive modeling approach proved more robust against the traditionally challenging data constraints in CVR estimation.

The authors also make notable contributions to the research community by releasing a public dataset with features critical for co-training CTR and CVR models. This dataset, capturing sequential dependencies between click and conversion events, provides a valuable resource for furthering large-scale investigations into multi-task modeling approaches.

The implications of this research extend well beyond the immediate improvements in CVR estimation. The model's ability to generalize over non-click events suggests potential applicability across various domains where user actions are sequentially dependent, such as digital advertising and content recommendation. Future explorations might enhance ESMM by incorporating additional stages of user interaction, such as request actions, thereby optimizing end-to-end conversion prediction.

Overall, this paper paves the way for more sophisticated models capable of capturing nuanced user behaviors through integrated multi-task learning. The publication serves as a critical touchstone for subsequent advancements in CVR modeling, offering a powerful alternative to classical predictive methods and highlighting the transformative impact of entire-space modeling strategies.