Conversion Rate Prediction Model For E-commerce Mobile Advertising Integrating User Features And Advertising Features

E-commerce companies usually use some types of mobile targeted advertising to help themselves to increase the visits and sales volume of the platform,such as short messages advertising and application push notifications.For them,the most important thing is the effect of advertising.The key factor that affects the effectiveness of targeted advertising is the accuracy of the audience.If frequent advertising is placed on uninterested and irrelevant users,not only will it not increase the visits and sales of the platform,but it will also trigger a series of advertising avoidance behaviors such as user complaints,unsubscribing,and closing APP message notifications,and even have a negative impact on the user’s attitude towards the company.With the continuous deepening of machine learning research and the emerging findings of artificial intelligence algorithms,big data models,and deep learning,a solution of predicting the conversion probability of users after receiving advertisements to improve the accuracy of advertising audience has become feasible.The key of the scheme is the accuracy of the purchase conversion rate prediction of e-commerce mobile advertising.This paper will study how to further improve the accuracy of e-commerce mobile advertising purchase conversion prediction on the basis of predecessors.This paper uses the real business data of e-commerce enterprises for research.In the aspect of feature extraction,two kinds of features of user and advertisement are selected,including user interaction behavior,user consumption behavior,user basic information and portrait label,and advertising content and advertising disturbance.In the aspect of prediction model,this paper selects nine classification prediction models from the traditional model,tree model and integrated model.By comparing the performance of the algorithm,the best model and feature combination are found.The experimental results show that the prediction effect of multi-dimensional feature combination is the best.Therefore,we should include the user’s basic information,user’s interaction behavior,user’s consumption behavior,advertising content and advertising disturbance into the model.And among the nine models,random forest algorithm has the best prediction effect.