O2O Coupons Personalized Recommendation System Based On Feature Engineering And Multiple Classifiers Fusion

In recent years,under the wave of "Internet +",various industries use data to empower and enhance the competitiveness of enterprises.In the emerging O2 O wave,various businesses have rich online and offline user data.The utilization of these data can enhance the competitiveness of merchants and improve the user experience.Attracting users through electronic coupons is one of the most cost-effective marketing methods in the O2 O industry.However,the rate of cancel after verification of coupons is very low when merchants blindly issue coupons,which seriously affects the marketing effect and challenges the marketing budget of merchants.If we can predict whether users will use coupons,and then issue the right coupons to the target users,we can improve the cancel after verification of coupons,marketing effect and user experience.The application of data engineering technology can be used to predict the effect of electronic coupons,optimize the issuing of coupons and improve the cancel after verification of coupons.In the field of e-commerce prediction task,feature engineering and algorithms are an indispensable part.At present,thousands of classification algorithms have been proposed and studied,but there is not much discussion on the content of feature engineering in the field of O2 O coupon.The main research contents of this thesis are as follows:1.In order to enrich the information of data and improve the prediction effect,feature construction and selection are carried out.In this thesis,a large number of features are constructed according to the original data,and they are grouped according to feature types.According to the impact of feature groups on the result,the importance of feature groups is evaluated and feature selection is carried out.2.In order to further improve the effect,this thesis carries out multi-model fusion.In the single model prediction,each algorithm has its own advantages.Through the multi-model fusion,the advantages of each algorithm are given full play to improve the prediction effect.3.Taking the above feature engineering and multi-model fusion as the core to design and realize the O2 O coupons personalized recommendation system.The data set of this thesis is derived from the real online and offline consumption data of Koubei users provided by tianchi O2 O coupons usage prediction contest,and the 0evaluation standard is the average AUC of the prediction of cancel after verification of coupons(The AUC value is the area under the ROC curve.The higher the AUC value,the better the prediction performance).In this thesis,AUC is improved to 0.80728 through feature construction and selection,and through multi-model fusion,AUC is further improved to 0.81120,and the prediction effect is significantly improved.Until now,the result ranks second among the 13,329 teams in the ranking of the AUC of the prediction of cancel after verification of coupons.Based on the above core methods,this thesis designs and realizes the O2 O Coupons Personalized Recommendation System,which truly realizes "invest in what they like".This system will improve the cancel after verification of coupons and enhance the competitiveness of merchants and user experience.