Several Kinds Of Advertisement Prediction Models Based On Empirical Bayes Method

The advertisement service via keywords bid ranking is the largest revenue source of search engine providers,and important channel of corporate branding and sales as well.In the process of keywords advertising,advertisers' main task is to manage the keywords and bid on the keywords,achieving the purpose of optimizing the allocation of limited budget and maximizing the advertising effect.How to predict the effect of keyword advertising is a hot and difficult research topic.Aiming at this situation,this study develops the prediction model of advertising conversion rate and sales ranking based on empirical Bayes method.In recent years,there has been a great deal of research on empirical Bayes,both in methodology and in application,and causing wide public concern.Empirical Bayes is a method to estimate some priori properties of unknown parameters by using existing data.The empirical Bayes method considers the priori information of parameters by regarding the parameters as random variables,then obtains the priori information of parameters by using the method of frequentist.In consideration of the excellent properties of empirical Bayes in the parameter estimation,this paper applies the Bayes method to prediction model of the advertising conversion rate and sales ranking.This paper also studied the keywords factors that affecting the advertising conversion rate from the advertisers' perspective,and has built the conversion rate model based on hierarchical Bayesian model.In this model,the Markoff Montecarlo method is used for sampling and calculating,and empirical Bayes method used for parameters estimation.The model had been applied to Baidu bidding data for 3 months.The empirical results show that:the nature of advertisement has a greater impact on the conversion rate,the length of the advertising keywords,trademark information,verbs and specific product information has a significant impact on the conversion rate.Higher ranking position also contributes higher conversion rate.The effect of advertising is ultimately reflected in the sales,this paper constructs a hierarchical Bayesian network advertising effect model to predict the corresponding product sales ranking.The model refers to the competition with other products,it constructs a competing model with time fading effect to be applied to the prediction of online sales ranking.Based on the data of mobile phone sales of Jingdong mall,we proved that the model with hierarchical Bayesian structure is better than the simplified model without random elements,by seizing the heterogeneous advertising responses of product and keywords.The results of this study have a certain impact on the prediction of online advertising effectiveness,and help advertisers to make decisions of advertising budget allocation effectively.