基于特征提取和机器学习的电商数据可视化分析系统设计

Design of e-commerce data visualization and analysis system based on feature extraction and machine learning

针对传统电商数据可视化分析系统对用户行为特征提取准确率低,导致用户行为自动识别效果不佳,无法进行电商数据可视化操作和数据关联分析的问题,提出基于特征提取和机器学习的电商数据可视化分析系统。系统通过Scrapy网络爬虫框架进行数据采集;采用TF-IDF权重法进行特征提取和特征向量空间生成;之后利用支持向量机SVM对特征进行分类,最后通过Django网络开发框架+JavaScript技术进行可视化系统实现。结果表明,TF-IDF算法生成向量空间维度为(5950,12530),明显优于其他特征提取策略。在不同的特征提取策略中,支持向量机SVM的精确率、召回率和F1值分别为97.55%、98.42%和96.34%,均高于朴素贝叶斯和逻辑回归分类模型。说明提出的算法和模型可对电商用户行为特征进行准确提取和行为分类识别。验证发现,本系统可进行电商数据陈列、图表展示和动态交互,满足电商数据可视化分析需求,进一步提升了电商数据可视化水平。

In view of the problem of traditional e-commerce data visual analysis system with low accuracy of user behavior feature extraction,resulting in low automatic identification effect of user behavior,and unable to conduct e-commerce data visual operation and data correlation analysis,an e-commerce data visual analysis system based on feature extraction and machine learning is designed.The system collects data through Scrapy network crawler framework;adopts TF-IDF weight method for feature extraction and feature vector space generation,classifies features through SVM,and finally implements visualization system through Django network development framework+JavaScript technology.The results show that the TF-IDF algorithm generates a vector space dimension of(5950,12530),which is significantly better than the other feature extraction strategies.Among the different feature extraction strategies,the exact rate,recall and F1 values of the SSVM were 97.55%,98.42%and 96.34%,respectively,which were higher than the naive Bayesian and logistic regression classification models.The proposed algorithm and model can accurately extract and classify the behavior characteristics of e-commerce users.The verification found that the system can carry out e-commerce data display,chart display and dynamic interaction,to meet the needs of e-commerce data visual analysis,and further improve the level of e-commerce data visualization.