Research On Ads' Click-Through Rates Predicting Based On Recurrent Neural Networks

With the explosive development of Internet advertising for more than a decade,Computational Advertising has emerged as the times requirement.As one of the most important way to calculate the flow of advertising,from search advertising to program trading,or mobile Internet native advertising,predict Advertising click-through rates(CTR)plays a key role in Computational Advertising.Advertising click-through rates is mainly based on massive user history data and under the complex orientation rules,with the help of big data technology and machine learning model,the candidate ads are sorted and predicted then show the right ads to the right audience.How to solve the ability of limited learning in linear model learning and the problem of relationship between advertising features is fully exploited in non – linear model,which has been arised the focus of research in related areas.On the premise of fully investigating the machine learning models commonly used to predict Advertising click-through rates,this paper proposed a kind of based on the Gated Recurrent Unit Neural Networks(GRU)model to solve the problem of predicting Advertising click-through rates.Further,by optimizing the step length method of t the gated recurrent unit neural networks,which makes the model in the iterative rounds of less to reach the optimal point better and faster,so as to improve the prediction ability of the model.The main work and achievements are as follows:(1)Carrying out the Feature Engineering is aimed at shallow learning model and deep learning model,include data analysis,preprocessing,feature selection and feature design and so on.This paper proposed an improved recurrent neural network – the gated recurrent unit neural networks,this networks replace hidden layers of general recurrent neural network gated unit structure,using the special gated unit structure to control the gradient propagation,so as to improve the learning ability of information characteristics.(2)On the base of the gated recurrent unit neural networks,the design of an optimal step length control method is proposed.First,set up a large step to search for the approximate global optimum faster,by small step index through the iterative attenuation to find local optimal,improve the iterative speed and accuracy of model.Finally,the AUC value was 0.05932 higher than the recurrent neural network and 0.003855 higher than the ated recurrent unit neural networks without optimization of step length control.(3)This paper compared the effects of six models,include logistic regression,naive bayes,random forest,recurrent neural networks,gated recurrent unit neural networks and its optimization of step length control version.The experiment results show that the model based on the gated recurrent unit neural networks and its optimization of step length control version has the best effect on the prediction of the Advertising click-through rates.