基于深度增强学习的个性化动态促销

Personalized Dynamic Promotion Based on Deep Reinforcement Learning

随着大数据技术和应用的普及,数据驱动的决策优化已成为企业科学管理的发展趋势。本文提出了一种基于深度增强学习的策略框架,整合企业领域知识解决收益管理中复杂的动态促销问题。基于一家在线生鲜零售商随机抽取的2012名匿名消费者自注册开始的长期消费行为(共计363946条消费记录),以及该零售商的历史收益管理策略数据,结合仿真方法,本文构建了增强学习环境下的个性化动态促销模型并进行检验。结果发现,本研究提出的新策略可以有效地提升企业面向市场的收益管理能力,提高企业约18%的长期收益。数据驱动的个性化促销实现了长短期促销策略的动态平衡,从而帮助企业以更低的投入和对顾客更少的干预实现更高的回报。研究结论对数智技术应用于企业收益管理优化和混合智能方法具有显著的理论和实践意义。

The advance in big data technology has made data-driven optimization a popular tool in operations management.The paper develops a deep reinforcement learning approach that integrates domain knowledge with a forward-looking perspective to solve the complex dynamic promoting problem in revenue management.We test the validity of the approach using a total of 363,946 consumption records from 2,012 randomly selected anonymous consumers from an online fresh food retailer.Simulation results demonstrate that our approach increases the firm's revenue by 18%compared with static strategies.The approach also dynamically balances long-term and short-term promoting strategies,helping the firm achieve higher returns with lower investment and less customer intervention.Our approach and findings have important implications for dynamic promoting and revenue management and the development of hybrid intelligence methods.