Scholarlink

<p>In cross-border fast-moving consumer goods (FMCG) e-commerce, there exists a substantial contradiction between the rapid iteration of new media marketing and the rigid response of the supply chain. This contradiction often leads to systematic problems such as stockouts of key products during promotion periods, overstock of long-tail products, and misallocation of cross-border logistics resources. To break through the limitations of existing collaborative models, this study proposes a multi-agent deep reinforcement learning framework based on the attention mechanism and meta-strategy modulation—Adaptive Spatio-Temporal Synergistic Deep Reinforcement Learning (AST-SDRL). The model designs a dual-agent architecture focusing on marketing decision-making and supply chain optimization. The marketing agent fuses multi-source new media traffic data and real-time order sequence features through a Spatio-Temporal Graph Convolutional Network (ST-GCN) to dynamically generate regionalized marketing actions. The supply chain agent introduces a sequence decision model with a gated recurrent attention mechanism, predicts demand based on marketing actions, real-time inventory, and in-transit information. It also makes plans for buying stuff, moving goods between warehouses, and choosing who handles the shipping. A meta-strategy modulator is innovatively introduced to improve the system’s generalization ability in unsteady environments. Through meta-learning of historical multi-task scenarios, it realizes fast online adaptation of the agent’s policy network parameters. Experimental results show that the proposed AST-SDRL model remarkably outperforms the comparison baselines in core performance indicators, with an order fulfillment rate of 96.8% and a comprehensive profit margin increased to 16.2%. This study provides an effective algorithm framework and decision support tool for solving the real-time collaboration problem between marketing and supply chain in cross-border e-commerce. Its excellent robustness and online adaptive ability have important practical significance for improving the efficiency of enterprises’ global operations.</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

Research and development of the new media marketing-supply chain collaborative adaptation model for cross-border fast-moving consumer goods e-commerce based on deep reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月25日

查看全文
  • Wenxuan Xu,
  • Jinfeng Xiang

摘要

In cross-border fast-moving consumer goods (FMCG) e-commerce, there exists a substantial contradiction between the rapid iteration of new media marketing and the rigid response of the supply chain. This contradiction often leads to systematic problems such as stockouts of key products during promotion periods, overstock of long-tail products, and misallocation of cross-border logistics resources. To break through the limitations of existing collaborative models, this study proposes a multi-agent deep reinforcement learning framework based on the attention mechanism and meta-strategy modulation—Adaptive Spatio-Temporal Synergistic Deep Reinforcement Learning (AST-SDRL). The model designs a dual-agent architecture focusing on marketing decision-making and supply chain optimization. The marketing agent fuses multi-source new media traffic data and real-time order sequence features through a Spatio-Temporal Graph Convolutional Network (ST-GCN) to dynamically generate regionalized marketing actions. The supply chain agent introduces a sequence decision model with a gated recurrent attention mechanism, predicts demand based on marketing actions, real-time inventory, and in-transit information. It also makes plans for buying stuff, moving goods between warehouses, and choosing who handles the shipping. A meta-strategy modulator is innovatively introduced to improve the system’s generalization ability in unsteady environments. Through meta-learning of historical multi-task scenarios, it realizes fast online adaptation of the agent’s policy network parameters. Experimental results show that the proposed AST-SDRL model remarkably outperforms the comparison baselines in core performance indicators, with an order fulfillment rate of 96.8% and a comprehensive profit margin increased to 16.2%. This study provides an effective algorithm framework and decision support tool for solving the real-time collaboration problem between marketing and supply chain in cross-border e-commerce. Its excellent robustness and online adaptive ability have important practical significance for improving the efficiency of enterprises’ global operations.