The Prediction Of Real-Time Bidding Conversion Rate Based On Nonparametric Delay

The rapid development of information technology,big data and advertising marketing has made the advertising mode which based on the Internet environment gradually enter the stage of refinement.The real-time bidding advertising model has became the sought of many Internet companies.The emergence of real-time bidding ads greatly simplifies the management efficiency of advertising campaigns and the process of information exchange.It can satisfy advertisers more accurate ad serving by refining users.This advertising model utilizes technologies and strategies such as big data analysis and indicators prediction to fine-tune the granularity of system processing to the display request for each advertisement.Therefore,a large number of researches on real-time bidding advertisements have emerged at domestic and foreign,mainly focusing on related optimization algorithms and prediction of key indicators of advertising.The ad conversion rate is one of the key performance indicators for evaluating the effectiveness of advertising.The accuracy of prediction model determines the input-output ratio of advertisers.Therefore,it is a work of great practical significance to make researches on prediction models of real-time bidding.This research based on the postgraduate innovation projects.The paper focuses on how to construct the prediction model of real-time bidding conversion rate and show users the most interesting advertising.Then the model can improve the possibility of product transactions and improve the input-output ratio of advertisers and improve user’s satisfaction with the network environment.The specific work mainly includes the following aspects:(1)The paper has defined the concept of the real-time bidding conversion rate and analyzed the impact characteristics of it.From the macroscopic and microscopic perspectives,combined with the real-time bidding advertising model,the paper gives the definition of real-time bidding conversion rate.Then it analyzes the main factors affecting the advertising conversion rate from three aspects: advertising attributes,user attributes and product attributes.The component analysis method has been used to select several most important features to provide a theoretical basis for constructing a real-time bidding conversion rate model.(2)The paper has constructed a real-time bidding conversion rate model.To construct the real-time bidding conversion rate model,the paper has considered the factors of intrinsic conversion rate,user’s conversion delay and ad conversion attribution mechanism.In order to deal with the problem of user conversion delay in real-time bidding ads,a nonparametric approach has been used to model the user’s conversion time distribution.And then applies this distribution into the probabilistic multi-contact attribution analysis.(3)The paper has established a conversion rate prediction model based on LSTM model.Using the built conversion rate model can calculate the conversion rate of all ad campaigns for each ad activity in a day,then build an LSTM model to predict the conversion of all campaigns in the future.The prediction can provides data support for advertisers to formulate budget allocations for different campaigns.(4)Finally,the work of the thesis is summarized,and the future research directions are prospected.The thesis also conducts simulation experiments through real-time bidding advertising data,and initially analyzes and verifies the effectiveness of the real-time bidding conversion rate model constructed in this paper.