Research On Accurate Delivery Of O2O Coupons Based On LightGBM Algorithm

Online To Offline is an e-commerce model that combines online activities with offline experiences.In the context of mobile Internet,the delivery of O2 O coupons becomes a major marketing tool,the platform of O2 O is constantly thinking about how to mine users' behavior characteristics and preferences.The accuracy of the delivery of coupons depends on user's needs,so we can spend lower marketing cost to invigorate old users and attract more new users.Owing to the difference between O2 O e-commerce and traditional e-commerce,O2 O has its own unique offline experience link.If the traditional e-commerce research is directly applied to the O2 O e-commerce model,there will be a situation of low accuracy and poor effect,failure to achieve the goal of precise marketing.Therefore,by analyzing the specific characteristics of the O2 O model and combining the traditional e-commerce research methods,building a prediction model for the O2 O ecommerce is the starting point and the foothold of this paper.The main purpose of this paper is to build a precise model for the delivery of O2 O coupons,to focus on the prediction of the use of O2 O coupons for users,and to distribute the coupons to the most likely users through the prediction results,so as to achieve precise delivery.The main work of this paper can be concluded as follows:(1)Combining with the theory of O2 O coupons and precise delivery,the influencing factors of the use of coupons are analyzed,which serves as the theoretical basis for feature construction;(2)After analyzing the original data,combined with the business logic,design and construct the features that can be directly applied to the actual business scenario,including 5 basic feature groups and 4 combined feature groups,with a total of 78 features.;(3)A new feature selection algorithm RFPS(Random Forest-Pearson-SBS)is proposed,and the features are selected by RFPS algorithm.Finally,52 features with high feature impact factors are obtained;(4)To solve the problem of positive and negative sample imbalance,a composite algorithm of LightGBM is proposed based on the Easy-Ensemble algorithm.——E-LightGBM is used to construct a precise model for the delivery of O2 O coupons.Good results have been achieved,which prove that the algorithm is suitable for wide promotion and application in similar problems;(5)Combining with the development of the O2 O industry and the experimental results,this paper puts forward some suggestions for the precise delivery of the O2 O coupons,and provides data support for the precise delivery strategy.The main conclusions drawn in this paper are as follows:(1)Under the background of the O2 O model,this paper proposes a prediction model for the use of coupons based on user historical data analysis,and provides theoretical basis and data support for the accurate delivery of O2 O coupons.(2)After using the RFPS algorithm for feature screening,the publicly-tested AUC value increased by 0.028,and the model runtime decreased by 16%,and the model is improved classification effect while reducing the complexity of the model;(3)Using the E-LightGBM algorithm to construct the O2 O coupon accurate delivery model,the publicly-tested AUC value is 0.798,which is superior to other classification integration algorithms in classification performance,which proves that the algorithm is more suitable for dealing with the data in the scenario of O2 O Ecommerce with unbalanced positive and negative samples and large amount of data.