Research On Prediction Method Of E-commerce User’s Purchase Behavior Based On Big Data Analysis

With the development of big data and Internet,mass users tend to shop online.Through the analysis of the relationship between these users and commodity information to explore their behavior habits and potential purchase demand is the focus of e-commerce enterprises.Therefore,this paper proposes a prediction method of e-commerce users’ repeat purchase behavior based on SAE stacking.The specific content and results are as follows:(1)In order to solve the problem of huge and messy behavior data of e-commerce users,a data processing method based on data equalization and visualization is proposed.Firstly,the original data set is cleaned,and the data is balanced by mixed sampling,and then the balanced data is normalized;secondly,the data is analyzed visually to further mine the internal rules of the data;finally,the feature set is constructed from the perspective of user,merchant and user merchant pair.(2)Aiming at the problem of poor generalization ability of manual feature extraction,this paper introduces deep learning algorithm and proposes a re purchase behavior prediction method of e-commerce users based on SAE xgboost.According to the reconstruction error and random gradient descent algorithm,a three-layer SAE feature extraction model is established,and xgboost algorithm is used to predict users’ repurchase behavior.Simulation results show that SAE xgboost has the highest prediction accuracy compared with xgboost and PCA xgboost.(3)Aiming at the problems of poor generalization performance and insufficient prediction accuracy of a single prediction model,a prediction model of e-commerce user repurchase behavior based on the Stacking integration method is proposed.The five-fold cross-validation method was used to perform heterogeneous fusion of the three classification models of LR,SVM and Xgboost.The simulation results show that:Compared with AE-SVM and SAE-Xgboost,the prediction accuracy of SAE-Stacking model is increased by 3%to 5%,which not only achieves higher prediction accuracy,but also has stronger generalization ability.In general,the research results of this topic have important significance and application value in the aspects of user behavior prediction,design of recommendation system,further realization of precision marketing and so on.