| In the era of big data,the Internet platform has many users’ historical data and click logs.The platform makes reasonable and full use of the data to realize personalized and accurate recommendation.This approach is of great significance to improve the user experience and increase the revenue of the platform.Click through rate prediction is an important part of online advertising and personalized accurate recommendation,how to accurately predict the click through rate and sort the advertising information is a very important research content.In the problem of click through rate prediction,efficient learning interactive features is the key to improve the prediction ability of the model.In this paper,we study the click through rate prediction algorithm based on feature interaction and attention mechanism,and propose a new prediction model to improve the accuracy of prediction.The main contents are as follows:Firstly,in this paper,we comb and analyze the research results of shallow and deep click through rate prediction models related to feature interaction and click through rate prediction models related to attention mechanism,and introduce the development of the model.Secondly,starting with the problems of insufficient interaction of low-order continuous features and limited expression of combined features in the traditional model,taking the Deep FM model as the basic framework,We innovatively introduce the gradient boosting decision tree(GBDT)and attention mechanism simultaneously,and propose a new click through rate prediction model based on different features interactive attention mechanism.The model can not only fully mine the low-order and high-order interactive information of features and enhance the expression of combined features,but also meet the functions of memory and generalization without artificial feature engineering.GBDT discretizes continuous features,fully excavates the original continuous feature information,realizes the feature interaction between low-order continuous features and low-order discrete features,the attention mechanism can efficiently express the relationship and importance between feature intersections,which solves the problem of limited expression of combined features of traditional models.Finally,two groups of experiments are designed on two data sets.One is the comparison experiment of benchmark model.The results show that the prediction effect of this model is better than the advanced benchmark model.The second is ablation experiment,which verifies that the two sub modules of GBDT and attention mechanism added at the same time have promoted the performance of the model.The two groups of experimental results show that the new model proposed in this paper can effectively improve the accuracy of click through rate prediction,and prove the value of this study. |
