Scholarlink

<p>Telecommunications operators face intensifying competition and evolving customer preferences, necessitating sophisticated strategies for value-added service (VAS) optimization. This paper proposes a novel multi-agent reinforcement learning (MARL) framework that jointly optimizes dynamic pricing and personalized recommendation strategies through coordinated agent interactions. The system architecture employs specialized agents managing pricing and recommendation decisions, integrated through attention-based communication protocols and multi-objective reward structures balancing revenue maximization with customer satisfaction. Centralized training with decentralized execution enables agents to leverage global state information during learning while maintaining scalable deployment. Experimental validation using real-world data from a major European telecommunications operator demonstrates substantial improvements: 23.7% revenue growth (95% CI: 21.3%–26.1%, <i>p</i> &lt; 0.001), 81.6% recommendation accuracy (± 2.1%), 4.8% point churn reduction (<i>p</i> &lt; 0.01), and 38% customer satisfaction enhancement compared to traditional approaches. These results remained statistically significant across ten independent experimental runs. Pilot deployment serving 1&#xa0;million subscribers confirms practical viability with sustained performance gains across diverse operational scenarios. The framework advances theoretical understanding of cooperative learning in commercial environments while providing actionable tools for telecommunications operators to enhance both profitability and customer experience.</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

Joint optimization of dynamic pricing and personalized recommendation for telecommunications value added services via multi agent reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Zhengyuan Zou,
  • Huaming Ling,
  • Chenchen Zhang,
  • Zewei Pan,
  • Xiaoqing Wang,
  • Jiayu Chen

摘要

Telecommunications operators face intensifying competition and evolving customer preferences, necessitating sophisticated strategies for value-added service (VAS) optimization. This paper proposes a novel multi-agent reinforcement learning (MARL) framework that jointly optimizes dynamic pricing and personalized recommendation strategies through coordinated agent interactions. The system architecture employs specialized agents managing pricing and recommendation decisions, integrated through attention-based communication protocols and multi-objective reward structures balancing revenue maximization with customer satisfaction. Centralized training with decentralized execution enables agents to leverage global state information during learning while maintaining scalable deployment. Experimental validation using real-world data from a major European telecommunications operator demonstrates substantial improvements: 23.7% revenue growth (95% CI: 21.3%–26.1%, p < 0.001), 81.6% recommendation accuracy (± 2.1%), 4.8% point churn reduction (p < 0.01), and 38% customer satisfaction enhancement compared to traditional approaches. These results remained statistically significant across ten independent experimental runs. Pilot deployment serving 1 million subscribers confirms practical viability with sustained performance gains across diverse operational scenarios. The framework advances theoretical understanding of cooperative learning in commercial environments while providing actionable tools for telecommunications operators to enhance both profitability and customer experience.