Design And Implementation Of Refinement Operation System For E-commerce Based On Classification Model

With the gradual saturation of the Internet market in all directions,users have become the resources that enterprises scramble for,and the importance of user growth is becoming stronger and stronger.Taking e-commerce enterprises as an example,major platforms usually spend a certain amount of cost and launch a variety of preferential policies to attract users to use app for consumption,so as to increase the number of users and ultimately achieve the goal of revenue growth.Under the premise of paying a huge cost,the enterprise will achieve short-term user scale growth effect.But after the end of preferential policies,enterprises will lose their new users again.The loss of new users makes it difficult for companies to meet their revenue growth targets.How to keep the new users on the platform for continuous consumption and make them change from new users to loyal users is a problem that many enterprises tend to ignore at present.Based on this background,this paper studies the new user retention of Internet e-commerce enterprises and designs a refinement operation system.The specific research contents are as follows:(1)This paper uses the half-year user consumption data of a famous Internet e-commerce platform as the basic data set.In addition,this paper analyzes the data from global and specific user perspectives,and preprocesses the missing values;(2)Through in-depth mining of the original data set,this paper carried out feature engineering from three perspectives of user,business and user-business association,and expanded the original 10 features to 110 features.In order to prevent the phenomenon of model overfitting,this paper uses the Random Forest(RF)algorithm to evaluate and screen the importance of features.(3)In this paper,two classical classification models,eXtreme Gradient Boosting(XGBoost)and Light Gradient Boosting Machine(LightGBM),were used as the basic classifier to build the prediction model of the probability that the user will consume again in the future,and the prediction results of the two basic classifiers were combined by the voting method.After the experiment,the AUC value of the fusion was 0.703011,which significantly improved the prediction effect compared with that of a single classifier.(4)In this paper,based on the prediction model of users' future consumption behavior,this paper designs and implements a refinement operation system based on Django framework.The main functions of the system are to divide the user group according to the probability of users' future consumption,to provide accurate operation tools and system authority management.This paper aims to solve the problem of revenue growth by mining the potential future re-consumers of new users in the field of Internet e-commerce,and to predict the probability of users' re-consumption in the future by mining the data of users' historical consumption behavior,so as to help enterprises find high-potential users among new users and conduct accurate operations.In the research field at home and abroad,there are few research results on the retention of new users.The research of this paper provides a solution,which can help enterprises to increase users efficiently and cost effectively in the fierce market competition.