摘要
为提高用户购买行为的预测性能,精准定位用户需求,实现线上商品的个性化推荐,提出了一种电商用户购买行为预测方法。该方法通过采集商品的历史交互数据,经数据清洗、构造消费者特征群、商品特征群和时间特征群三大衍生特征群、利用特征相似性分析进行特征选择,采用极限梯度提升树(eXtreme Gradient Boosting,XGBoost)建立用户购买行为预测模型,并利用遗传算法(Genetic Algorithm,GA)对模型的超参数进行优化(GA-XGBoost)。实验结果表明,与逻辑回归、决策树等传统机器学习方法和单一的XGBoost模型相比,GA-XGBoost模型的预测精度明显提升。
In order to improve the predictive performance of user purchasing behavior,accurately locate user needs,and realize personalized recommendation of online goods,an e-commerce user purchasing behavior prediction method was proposed.The method includes several steps as follows:collecting historical interaction data of commodities,data cleaning,constructing three derivative feature groups,namely consumer feature group,commodity feature group and time feature group,making feature selection by feature similarity analysis,and optimizing the hyperparameters of the model(GA-XGBoost)by adopting eXtreme Gradient Boosting(XGBoost)and Genetic Algorithm(GA).Experimental results show that compared with traditional machine learning methods such as logistic regression and decision tree and single XGBoost model,the prediction accuracy of GA-XGBoost model is significantly improved.
作者
吴鑫
李君
茅智慧
吴耀辉
郑李园
WU Xin;LI Jun;MAO Zhihui;WU Yaohui;ZHENG Liyuan(Zhejiang Wanli University,Ningbo Zhejiang 315100)
出处
《浙江万里学院学报》
2022年第4期86-92,共7页
Journal of Zhejiang Wanli University
基金
浙江省教育厅一般科研项目(Y202045452)
大学生创新创业训练计划项目(S202010876094,202010876035)
浙江省自然科学基金项目(LY18F010001)
宁波市自然科学基金项目(2019A610076)。
