基于用户行为序列的电商需求销量预估研究

Research on E-Commerce Demand and Sales Forecast Based on User Behavior Sequences

摘要: 在电商场景下精准预估供应链销量，对于优化冷链配置、提升配送效率与降低库存周转成本具有重要的现实意义。本文基于淘宝开源电商数据集，突破传统仅依赖历史销量的局限，将用户在线行为(点击、加购、收藏)纳入销量预估框架。研究发现，VAR模型有效地揭示了流量转化过程中各个层级的动态传导机制，而Transformer模型则精准捕捉了由大促活动引发的非线性需求脉冲。二者结合提升了短期订单量的预估精度。经验证，VAR模型能有效捕捉变量间动态联动关系，Transformer模型在非线性拟合上表现更优，二者形成互补。模型结果显示，“双十二”类促销活动能有效促进电商市场发展，相关企业应提前布局应对物流高峰。本文旨在为电商平台与物流企业探索轻量化、可落地的需求预估方案，为供应链协同决策提供数据支撑。

Abstract: Accurately predicting sales volume and supply chain throughput in e-commerce scenarios is of great practical significance for optimizing cold-chain allocation, improving distribution efficiency, and reducing inventory and turnover costs. Based on an open-source Taobao e-commerce dataset, this study breaks through the limitations of traditional sales forecasting by incorporating users’ online behaviors (clicks, add-to-cart, favorites) into the forecasting framework. It is found that the VAR model effectively reveals the dynamic transmission mechanism across all levels in the traffic conversion process, while the Transformer model accurately captures nonlinear demand shocks caused by livestreaming e-commerce or major promotion campaigns. Their combination improves the accuracy of short-term order volume prediction, providing a data-driven decision paradigm for omni-channel inventory allocation and front warehouse operations of e-commerce commodities. Verification shows that the VAR model effectively captures the dynamic linkage between variables, whereas the Transformer model performs better in nonlinear fitting, and the two models are complementary. The results indicate that shopping carnivals such as Double 12 effectively promote the positive development of the e-commerce market. Relevant enterprises and authorities should make advance arrangements to cope with logistics peaks during promotions. This study provides a lightweight and implementable solution for demand and sales forecasting for e-commerce platforms and logistics enterprises, and also offers data support for collaborative decision-making in the e-commerce supply chain.