Research On Forecasting Method Of Advertising Click Rate Based On Wide And Dien Model

With the explosive growth of Internet big data and the upgrade of computer technology,the advertising industry has brought tremendous changes.The prominent change occurred on the carrier of advertising communication.The traditional advertising is based on paper media.Now advertising is more turned to the Internet online platform,so the discipline of computing advertising also came into being.The generation of computing ads has brought huge business opportunities and benefits to Internet advertising platforms.Its core billing model is CPC(pay-per-click).Under this billing model,the prediction of ad click rate has become a hot spot and a core The problem has aroused the attention of industry and academia.Because increasing the accuracy of ad click rate prediction is equivalent to improving advertisers,improving Internet user experience satisfaction,increasing Internet platform revenue and user stickiness.(1)This article first selects samples from Amazon's public data set,selects a subset of electronic products in the data set for experiments,and then generates a training set and a test set based on the user's click sequence.The training set contains positive and negative samples.(2)This article conducts an experimental analysis of the advertisement clickthrough rate prediction model.The deep neural network model has strong learning ability,does not require artificial feature combinations,and learns high-order nonlinear features autonomously,but it will overgeneralize when the data amount is insufficient.The breadth depth training model trains the linear model and the deep neural network model jointly,which not only has memory ability but also generalization ability.The deep interest network introduces a local activation mechanism in the deep neural network structure,and assigns weight to the user's historical information by analyzing the correlation between the advertising commodity and the user's historical behavior,which improves the accuracy of the click rate prediction.(3)After experimental research and analysis,this paper proposes the Wide and Dien model for ad click rate prediction.In this model,the Deep Interest Evolution Network(Deep Interest Evolution Network,DIEN)can extract the interests of users and arrange them into a time series according to time.By introducing the GRU module to find the implicit dependencies between user interests,the evolution trend of interest is formed.Wide side joint training.Finally,through the experimental comparison with the baseline model,this paper verifies that the model in this paper has significantly improved the prediction effect.(4)This article implements a cultural tourism commodity recommendation platform.The core recommended commodity module of the platform uses the Wide and Dien model proposed in this article.