Scholarlink

<p>Global e-commerce is projected to exceed USD 8.03 trillion by 2025, intensifying the demand for intelligent systems that can integrate and optimize diverse digital marketing strategies. Traditional studies often examine Search Engine Optimization (SEO), Social Media Marketing (SMM), Pay-Per-Click (PPC), and email campaigns separately, thereby limiting actionable insights. To address this gap, we propose a novel cross-channel transformer framework that unifies heterogeneous marketing data streams within an intelligent system for improved customer acquisition and conversion. The architecture extends the transformer paradigm by incorporating multi-head self-attention with hierarchical channel encoders to separately model SEO, SMM, PPC, and email signals, followed by a cross-channel fusion layer that learns interdependencies between platforms. A contextual embedding module captures temporal engagement dynamics, while a conversion-focused decoder outputs probabilistic predictions for customer actions. Methodologically, preprocessing steps include categorical encoding, normalization, and class balance adjustment, ensuring model robustness. Evaluation employs key quantitative metrics—Click-Through Rate (CTR), engagement, and conversion rate—defined with explicit thresholds for interpretability. Furthermore, interpretability is enhanced through attention heatmaps at the channel and feature level, complemented by SHAP-based attribution to explain individual predictions. The proposed Cross-Channel Transformer achieved the highest predictive performance across all tasks, reaching 90.8% CTR accuracy, 86.3% engagement prediction, and 84.7% conversion accuracy, outperforming all baseline models. SHAP analyses further confirmed strong channel-level contributions, with SMM and SEO emerging as the most influential predictors of customer engagement and conversion. This work advances the integration of intelligent systems in e-commerce by introducing a novel transformer architecture that simultaneously captures intra-channel representations and cross-channel synergies, offering a scalable foundation for future personalization and adaptive marketing strategies.</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

Intelligent Systems for E-Commerce: A Cross-Channel Transformer Framework for Customer Acquisition and Conversion Optimization

  • 期刊论文

  • 发表日期：

    2025年12月18日

查看全文
  • Lama Khoshaim

摘要

Global e-commerce is projected to exceed USD 8.03 trillion by 2025, intensifying the demand for intelligent systems that can integrate and optimize diverse digital marketing strategies. Traditional studies often examine Search Engine Optimization (SEO), Social Media Marketing (SMM), Pay-Per-Click (PPC), and email campaigns separately, thereby limiting actionable insights. To address this gap, we propose a novel cross-channel transformer framework that unifies heterogeneous marketing data streams within an intelligent system for improved customer acquisition and conversion. The architecture extends the transformer paradigm by incorporating multi-head self-attention with hierarchical channel encoders to separately model SEO, SMM, PPC, and email signals, followed by a cross-channel fusion layer that learns interdependencies between platforms. A contextual embedding module captures temporal engagement dynamics, while a conversion-focused decoder outputs probabilistic predictions for customer actions. Methodologically, preprocessing steps include categorical encoding, normalization, and class balance adjustment, ensuring model robustness. Evaluation employs key quantitative metrics—Click-Through Rate (CTR), engagement, and conversion rate—defined with explicit thresholds for interpretability. Furthermore, interpretability is enhanced through attention heatmaps at the channel and feature level, complemented by SHAP-based attribution to explain individual predictions. The proposed Cross-Channel Transformer achieved the highest predictive performance across all tasks, reaching 90.8% CTR accuracy, 86.3% engagement prediction, and 84.7% conversion accuracy, outperforming all baseline models. SHAP analyses further confirmed strong channel-level contributions, with SMM and SEO emerging as the most influential predictors of customer engagement and conversion. This work advances the integration of intelligent systems in e-commerce by introducing a novel transformer architecture that simultaneously captures intra-channel representations and cross-channel synergies, offering a scalable foundation for future personalization and adaptive marketing strategies.