Scholarlink

<p>The optimization of e-commerce advertising campaigns increasingly relies on machine learning models. However, these models often function as ‘black boxes,’ lacking the transparency needed for marketers to make informed, risk-assessed decisions about budget allocation and bidding strategies. This critical gap limits both the effectiveness of and trust in automated advertising systems. This paper addresses this challenge by proposing a novel framework that embeds explainability directly into the optimization process. This research introduces two key innovations: (1) the application of feature attribution techniques, including Integrated Gradients, to demystify the inner workings of profitability prediction models; and (2) the development of a probabilistic decision-making framework using Bayesian Decision Trees to generate transparent, actionable insights. The proposed framework translates complex probabilistic outputs into intuitive decision rules, revealing the specific factors that drive campaign success. It offers practical, data-driven strategies for improving e-commerce ad performance and supports more informed, confident decision-making. The framework enables adaptive bidding strategies, optimized budget allocation, and improved decision-making in e-commerce advertising.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable ML Models for E-commerce Ad Optimization

  • 期刊论文

  • 发表日期：

    2025年12月24日

查看全文
  • Arti Jha,
  • Pratyut Sharma,
  • Kamlesh Tiwari

摘要

The optimization of e-commerce advertising campaigns increasingly relies on machine learning models. However, these models often function as ‘black boxes,’ lacking the transparency needed for marketers to make informed, risk-assessed decisions about budget allocation and bidding strategies. This critical gap limits both the effectiveness of and trust in automated advertising systems. This paper addresses this challenge by proposing a novel framework that embeds explainability directly into the optimization process. This research introduces two key innovations: (1) the application of feature attribution techniques, including Integrated Gradients, to demystify the inner workings of profitability prediction models; and (2) the development of a probabilistic decision-making framework using Bayesian Decision Trees to generate transparent, actionable insights. The proposed framework translates complex probabilistic outputs into intuitive decision rules, revealing the specific factors that drive campaign success. It offers practical, data-driven strategies for improving e-commerce ad performance and supports more informed, confident decision-making. The framework enables adaptive bidding strategies, optimized budget allocation, and improved decision-making in e-commerce advertising.