多行为序列建模下的电商用户购买行为与关键影响因素分析

Analysis of E-Commerce User Purchase Behavior and Key Influencing Factors under Multi-Behavior Sequence Modeling

摘要: 在电商领域持续扩张的背景下，用户在平台上的多样化互动为构建推荐模型和开展购买意愿预测提供了充足的数据基础。本研究利用阿里天池发布的电商行为数据作为分析对象，构建融合浏览、关注等多类行为特征的序列化模型，并采用XGBoost算法对用户的实际购买动作进行预测。研究中进一步引入SHAP方法，以增强模型的解释框架，从量化视角刻画不同特征的重要程度及其潜在影响路径。与Logistic Regression等传统对照模型相比，XGBoost在Accuracy、AUC与F1等核心指标上均表现更为突出，体现出较强的预测能力。通过SHAP的解释结果可以观察到，总浏览量等长期且累积性的行为变量对购买决策具有显著影响；而短期行为以及收藏类动作，对用户最终是否下单的贡献度相对较弱。

Abstract: In the context of the rapid development of the e-commerce industry, the diverse and abundant user behaviors have laid a solid data foundation for building recommendation systems and predicting purchasing behavior. This study utilizes e-commerce user behavior data provided by Alibaba Tianchi as the research material and constructs a sequential model that integrates multiple behavioral features such as browsing and collecting. The XGBoost algorithm is employed to predict purchasing behavior, while the SHAP method is introduced to enhance model interpretability and quantitatively analyze the importance and impact of each feature. The results show that compared with baseline algorithms such as Logistic Regression, XGBoost demonstrates superior performance across multiple evaluation metrics—including Accuracy, AUC, and F1-score—indicating strong predictive capability. According to the SHAP analysis, factors such as the total number of browsing actions are key determinants of purchasing behavior, whereas short-term behaviors and collection actions contribute relatively less to explaining purchasing decisions.