Research And Application Of Advertising Click-through Rate Prediction Based On GBDT And Attention Mechanism

The core issue of computational advertising-click-through rate prediction,refers to predicting whether users will click on an advertisement.It is a key link in the profitability of Internet companies,and it is also a key research direction in industry and academia.At the same time,with the development of deep learning technology,many research institutions have also begun to apply deep learning technology in the click-through rate prediction model.In this paper,through in-depth research and summary of the existing clickthrough rate prediction models,it is found that both the traditional machine learning model and the deep learning model have certain deficiencies in feature interaction,especially the lack of consideration of the importance of different features.In response to the above problems,this paper innovatively proposes a model that integrates GBDT,Deep FM,and attention mechanism,that is,the GBDT-Deep AFM model.The effect of the model is verified by using the massive online advertising click-through rate data.The main work of this paper is as follows:(1)This paper has carried out data exploration and feature engineering on the used data set,charted and analyzed the characteristics of the data set and the statistical situation of some features,and based on business understanding,derived feature construction and statistical type of the data set feature construction.(2)In view of the lack of consideration of the importance of features in the feature interaction of existing models,this paper proposes the GBDT-Deep AFM model,which adds an attention mechanism to the Deep FM model,so that the model can take into account different features on the model prediction results At the same time,the GBDT model is added.The final model can combine the advantages of each sub-model to better mine the deep information in the data.Through the design and implementation of offline experiments,the results show that the GBDT-Deep AFM model has good performance in AUC and Logloss,has good learning and generalization capabilities,and has good prediction results on the data set.(3)Finally,according to the GBDT-Deep AFM model,a product recommendation system is designed and implemented,that is,the model is applied,and the system is used to display the recommendation effect of the model more intuitively.