Research On Precision Prediction Model Of Advertising Placement Based On GRU Neural Network

In recent years,with the rapid development and application of search engines,search engine-based advertising has become the largest form of online advertising,which shows great development prospects and commercial value.Click-through rate is an important criterion for judging the effectiveness of search engine advertising and the main basis for advertisers to advertise.Therefore,it is very necessary and realistic to study the click rate prediction of search engine advertisements.In the existing research,the main idea is to simulate user behavior based on the user's perspective,but this does not provide an effective basis for advertisers to advertise.In addition,most of the existing prediction algorithms adopt features that are isolated,and do not consider the relationship between data features and the influence of dependencies on predicted values.The research goal of this paper is to improve the intrinsic relationship between sequence data and mining features through the cyclical neural network from the perspective of advertisers,so as to improve the effect of click rate prediction and provide an effective basis for advertisers to advertise.The main research contents of this paper are as follows:Firstly,data analysis and feature processing are carried out.Sequential data is the expression of the dependence between data.This paper analyses the historical clickthrough rate of missing data sets from the perspective of advertisers,and finds that clicks in a period of time after an advertisement depend to a certain extent on the clickthrough study in a period of time before,and constructs time-based advertising sequence data which is suitable for training a series of algorithms of cyclic neural networks according to the dependency relationship;and then extracts features from the data.The feature data that is valid for the model is obtained.In order to solve the problem that one-hot encoding of text information produces huge feature dimension and isolation between encoding features,a CBOW model is proposed to encoding text features.Secondly,construct a click-through rate prediction model and experimental analysis.In this paper,the logistic algorithm is firstly improved based on the penalty term,and the Elastic Net-logistic model has better dimensionality reduction effect and better prediction effect.Then according to the shallow model's shortcomings of relationships between features and dependencies between data,the ad click rate prediction model based on GRU neural network is proposed,and the experimental comparison shows that the GRU prediction model is compared with the LSTM and the traditional BP and Elastic Net-logistic prediction models of the same cyclic neural network.There is a better prediction effect;finally,based on the respective advantages of Elastic Net-logistic and GRU models,a fusion prediction model is proposed to further improve the effect of click rate prediction.Thirdly,Design and implement a click-through rate prediction platform.Based on the above-mentioned click-through rate prediction model,the search engine advertisement click-through rate prediction platform is designed and implemented to provide an appropriate platform for advertisers to make advertising decisions.