摘要
以Spark软件为工具,对电商网站用户行为分析预测系统进行探讨,在此基础上,对一家电商平台进行了一系列的预处理,包括在一定的时间段内,对用户的行为进行处理,提出时间序列规则处理原始数据动态滑动窗口。用户行为分析实验表明,XGBoost的训练模式表现最好,而决策树的学习效果最差。XGBoost模型无需对全部的训练进行集合,是通过XGBoost在每一个滑行窗口内使用XGBoost来输出最后的预测。XGBoost在预测结果正确率、稳定性方面均较好。Spark平台主要由数据读入、RDD的创建、用户行为预测计算三部分构。相比Hadoop平台,基于Spark平台系统效率提高了近8倍,系统运行速度降低幅度较大,减少了电商网站运营成本,Spark平台系统可靠性较高。
Based on Spark software,this paper studies the user behavior analysis and prediction system of e-commerce websites,and on this basis,a series of pre-processing is carried out on an e-commerce platform,including processing user behavior within a certain period of time,and proposing time series rules to process the dynamic sliding window of raw data.User behavior analysis experiments show that XGBoost’s training mode performs the best,while decision trees have the worst learning effect.The XGBoost model does not need to set all the training,and uses XGBoost in each gliding window to output the final prediction.XGBoost has good accuracy and stability in prediction results.The Spark platform is mainly composed of three parts:data reading,RDD creation,and user behavior prediction and calculation.Compared with Hadoop platform,the system efficiency based on Spark platform has increased by nearly eight times,the system running speed has been greatly reduced,and the operating cost of e-commerce websites has been reduced.The system reliability of Spark platform is high.
作者
谢鑫
XIE Xin(Zhangzhou Vocational and Technical College,Zhangzhou 363000,Fujian,China)
出处
《贵阳学院学报（自然科学版）》
2023年第1期9-13,共5页
Journal of Guiyang University：Natural Sciences
