基于改进XGBoost算法的电商平台用户复购行为预测研究  

Research on repurchase behavior prediction of e-commerce platform users based on improved XGBoost algorithm

作　　者：叶昊 袁凯骏 沈伟 郑志强[3] 张宏俊 YE Hao;YUAN Kaijun;SHEN Wei;ZHENG Zhiqiang;ZHANG Hongjun(Zhongbo Information Technology Research Institute Co.,Ltd.,Nanjing 210012,China;Faculty of Engineering,University of Bologna,Bologna 40121,Italy;College of Modern Postal Services,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;China Communications Services Corporation Limited,Beijing 100073,China)

机构地区：[1]中博信息技术研究院有限公司,江苏南京210012 [2]博洛尼亚大学工程学院,意大利博洛尼亚401213 [3]南京邮电大学现代邮政学院,江苏南京210003 [4]中国通信服务股份有限公司,北京100073

出　　处：《通信与信息技术》2025年第2期11-16,共6页Communication & Information Technology

基　　金：国家自然科学基金(项目编号:61972208);江苏省农业科技创新基金(项目编号:CX(22)1007);江苏省研究生科研与实践创新计划项目(项目编号:KYCX22_1027,KYCX23_1087,SJCX24_0339,SJCX24_0346);南京邮电大学大学生创新训练计划项目(项目编号:XZD2019116,XYB2019331)。

摘　　要：针对电商平台的用户复购预测问题,提出了一种基于XGBoost算法的解决方案。研究通过深入分析用户行为数据,识别出影响复购行为的关键因素,并构建了逻辑回归与XGBoost模型进行对比分析。实验结果显示,初始XGBoost模型在测试集上虽表现稳定,但召回率较低,提示模型存在拟合不足或数据质量问题。通过细致的参数调优,如增加弱学习器数量、调整树的最大深度及降低学习率,模型性能得到显著提升,错误率由32.3%降至28.7%,准确率提升至71.3%,显示出XGBoost在复杂数据集上的预测优势。特征重要性分析揭示,用户浏览商品次数、重复购买同一商品的频次及点击率是预测复购的重要特征,而关注品牌数量及添加购物车比例的影响力较小。本文为天猫复购预测提供了有效的模型构建与优化策略,通过特征分析为电商平台提供了数据驱动的决策支持,对促进精准营销和增强用户粘性具有实践意义。Aiming at the problem of user repurchase prediction of e-commerce platform,this paper proposes a solution based on XG⁃Boost algorithm.Through in-depth analysis of user behavior data,the key factors affecting repurchase behavior are identified,and logistic regression model and XGBoost model are constructed for comparative analysis.The experimental results show that although the initial XGBoost model performs stably on the test set,the recall rate is low,suggesting that the model has insufficient fitting or data quality prob⁃lems.Through careful parameter tuning,such as increasing the number of weak learners,adjusting the maximum depth of the tree,and re⁃ducing the learning rate,the model performance has been significantly improved,the error rate has been reduced from 32.3%to 28.7%,and the accuracy rate has been increased to 71.3%,showing the predictive advantages of XGBoost on complex data sets.Feature impor⁃tance analysis reveals that the number of times users browse products,the frequency of repeated purchases of the same product and the click-through rate are important features to predict repurchase,while paying attention to the number of brands and the proportion of shop⁃ping carts have less influence.This paper provides an effective model construction and optimization strategy for Tmall repurchase predic⁃tion,and provides data-driven decision support for e-commerce platform through feature analysis,which has practical significance for promoting precision marketing and enhancing user stickiness.

关 键 词：XGBoost算法 复购预测 用户行为分析 数据驱动决策 电子商务 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]