基于社会学习行为的动态定价策略研究

With the rapid development of social networks, consumers’ social learning behavior has become a key factor that companies can't ignore when pricing. However, the curent literetures on pricing with consumers’ learning behavior mostly assume that learning behavior is consistent with the rules of Bayesian updating and is completely rational. The latest empirical results show that the social learning behavior of policymakers is not completely rational, but presents characteristics such as locality, limited memory, and limited trust. Therefore, based on the latest theoretical and empirical research, this project constructs three dynamic pricing models withbounded rational social learning: dynamic pricing with individual differences and time varying weights of social learning, dynamic pricing based on limited memory and local information social learning, as well as dynamic pricing under limited trust social learning. At the same time, we will further use social networks, opinion dynamics, dynamic programming and stochastic approximation theory to analyze and solve these models. Finally, we will explore data-driven methods to test the efficiency of our models and study its application strategy. This project belongs to the cross research areas of behavioral science, revenue management and logistics, and has important theoretical significance. Meanwhile, it can provide reference for enterprises in the Internet environment to formulate reasonable pricing strategies, and has broad application value.

随着社交网络的快速发展，消费者的社会学习行为已经成为企业制定价格策略时不可忽视的因素。而现有考虑消费者社会学习行为的定价研究大都假设学习行为是完全理性的，即符合贝叶斯更新规律。最新实证研究显示，消费者的社会学习并非完全理性，而是呈现局部性、非完美记忆、单调性、妥协性等有限理性特征。因此，本项目将以社交网络、观点动力学领域最新理论和实证研究为基础，构建几类基于有限理性社会学习的动态定价模型：基于个体差异和时变权重社会学习的动态定价，基于有限记忆和局部信息社会学习的动态定价，以及基于有限信任社会学习的动态定价等。同时，结合随机近似等领域新方法分析求解这些模型。最后通过数据驱动的方法完成模型的有效性检验，并对其应用策略进行研究。本项目属于行为科学、计算机科学与收益管理、供应链管理的交叉研究领域，研究内容具有重要理论意义，同时又可为互联网环境下的企业制定合理的定价策略提供参考，具有广阔的应用价值。

本项目基于社交网络和观点动力学等领域最新研究结果，构建了几种基于有限理性社会学习的动态定价模型，并提出了求解这些模型的算法，丰富了收益管理的理论研究和应用实践。近4年来，本项目团队展开的主要研究工作包括：（1）建立了基于有限理性社会学习的动态定价模型；（2）探索了基于强化学习和深度学习的动态规划算法；（3）提出了具有通用性的近似动态规划算法；（4）研究了消费者多行为因素下的动态定价策略。在该基金的资助下，本项目目前共发表8篇文章，其中SCI/SSCI检索4篇，基金委认定的重要中文期刊3篇。在培养人才方面，项目组协助培养博士7名，硕士生16名。同时，项目组经常资助项目团队出去交流，共参与13次国际国内重大会议，并与国外知名学者进行了深入的交流与密切的合作。累积资助5名博士前往境外知名高校进行联合培养。通过对项目课题的研究，整个项目团队提升了学术研究水平，开拓了研究的视野，锻炼和培养了人才，为研究有限理性假设和考虑消费者行为因素的定价问题提供了参考，为今后开展更加高水平的工作奠定了基础。