Research On The Conversion Rate Of O2O Coupons Based On Machine Learning

With the rapid development of communication technology,mobile network and broadband network quickly spread to every household,which greatly promotes the development of e-commerce.At present,new e-commerce based on big data and cloud computing emerges at the right moment.The business model O2O(Online To Offline),which combines Online and Offline channels organically,develops rapidly,and various marketing means emerge in endlessly,such as coupons.Analyzing the customer flow of O2 O coupon conversion and realizing fast and accurate prediction of whether users use the coupons they receive within the specified time can not only give merchants stronger sales ability and help them effectively release coupons,but also enable consumers with certain preferences to get real benefits.In this paper,the research on the conversion rate of O2 O coupons mainly includes the following three aspects:(1)Based on Tianchi Open Data Set,the user historical data of an e-commerce platform from January 1,2016 to June 30,2016 was analyzed,including the time of user's voucher collection,coupon discount,the nearest distance between user and offline merchant and the time of user's consumption in a merchant.Through the data visualization analysis of these dimensions,the potential law of data is excavated,and 58 features are constructed.They include six dimensions: user,merchant,coupon,user-merchant,user-coupon and other factors.(2)Based on the constructed feature set,four single models including random forest,GBDT,XGBoost and Light GBM,and stacking fusion model were used to predict the usage rate of O2 O coupon.Experiments verify the validity and rationality of the algorithm.(3)Considering that users have similar consumption psychology and merchants have similar marketing strategies,this paper further performs k-means clustering on users and merchants,classifies users into three categories and merchants into four categories,and then conducts stacking fusion model prediction.Experiments show that the accuracy of the ensemble model based on clustering algorithm has been improved.