| With the development of internet,the traditional advertising industry is shifting to the emerging market in internet.CTR(Click-Through-Rate)prediction which is concerned to the revenue of internet enterprises becomes one of today's hottest new technology markets.It's also the important application of scholarly research in industry.Faced the big data of user behavior,how to dig the data to meet the preference of user becomes more and more important.This paper focus on the actual application,using the user attributes,advertising information and user behavior then cutting in wide and deep model,puts out a CTR algorithm based on Wide&Deep model to meet different users' preferences.The main work of this paper includes the following aspects:(1)Reviewed the current research status of advertising CTR prediction,and analyzed their advantages and shortcomings.(2)Data pre-processing for CTR predicting : This paper analyzes the problems of the original data of CTR,and proposes a complete set of process for CTR prediction,including data cleaning,data sampling,data protocol and data conversion for collected user attributes,advertising information,user historical behavior,(3)Based on Wide&Deep's advertising click rate estimation algorithm: Based on the Wide&Deep model proposed by Google,combined with the Deep-FM algorithm and the timing characteristics of the CTR data set,this paper designs a factorization machine model based on user behavior.This model is divided into four parts,including the Wide layer,the Deep layer,the sharing Embedding and input layer.(4)Finally,this paper compares for various model to verify the effectiveness of the algorithm.Compared with the traditional click-rate estimation algorithm,the RD-FM model'experimental results show that the proposed algorithm has better performance than the traditional click-rate estimation algorithm on the AUC and RMSE.The algorithm fully exploits the advantages of the factorization machine model and the time series depth model. |
PDF Full Text Request