Research On Optimization Of Advertising Conversion Effect Based On Learning Method

With the development of the Internet and online advertising,advertisers have continuously improved their requirements for online advertising.Their advertising behavior has become more rational,and they are increasingly pursuing a payment method based on advertising conversion.However,problems such as large deviations in the prediction of converted users,high latency in the calculation of conversion results,and cold start by advertisers have caused the conversion effect of advertising to be unsatisfactory.In order to improve the conversion effect of advertising,this article designs an optimization model of advertising conversion effect based on learning methods,as follows:1.Designed an advertising exposure conversion rate prediction model based on RF+Deep FM.This article proposes the conversion rate of advertising exposure to replace the conversion rate of advertising to avoid the impact of data sparseness and sample selection bias on the prediction of converted users.At the same time,an algorithm model based on Random Forest(RF)and Deep FM is designed to accurately predict the conversion rate of advertising exposure,and further improve the accuracy of the prediction of advertising conversion users.2.Designed a real-time edge computing model based on computing migration.This paper designs a real-time edge computing architecture of device terminal-compute node-data center to improve the real-time performance and reliability of the calculation of advertising conversion results.And on the basis of the edge computing architecture,computing migration is used to further reduce the computing delay of advertising conversion results.At the same time,the computing migration model can also effectively reduce the load on the computing nodes of the advertising platform.Lowlatency advertising conversion result calculation can determine the effect of advertising conversion in real time and avoid invalid advertising.3.Designed a cross-domain recommendation method based on user interest transfer learning.Advertisers enter the advertiser's application through advertisements.Due to the lack of historical behavioral data,the advertiser's recommendation system cannot make personalized recommendations for them,causing some users to lose without effective conversion,that is,users cold start.To this end,this paper designs a method of migration learning of user interest,by migrating the user interest of the advertising platform to solve the problem of advertisers' user cold start,so as to promote the conversion of advertising users and improve the conversion effect of advertising.Experiments have proved that the conversion rate of advertising exposure proposed in this article is better than the conversion rate of advertisements in predicting converted users.Specifically,the AUC value increased by 0.0730 and the Logloss value decreased by 0.0614.At the same time,the RF+Deep FM algorithm model designed in this paper also has a good performance in converting user predictions compared to a single Deep FM model.The AUC value is increased by 0.0218,and the Logloss value is reduced by 0.0430.The calculation model of advertising conversion results based on calculation migration reduces the calculation delay by 0.86 s.In addition,it also effectively reduces the service load of the advertising platform's computing nodes.Cross-domain recommendation through user interest transfer learning can effectively solve the problem of advertisers' user cold start and promote the conversion of advertisers.Experiments show that this cross-domain recommendation method increases the download and registration rate of advertisers' applications by 4.3%.