Research On Online Advertising Click-Through Rate Estimation Based On Deep Learning

Online advertising is one of the two major sources of income for the Internet industry.In recent years,with the popularization of smartphones and then Internet,its market scale has been expanding,expected to exceed the trillion mark in 2022.For such a huge market,any strategic optimization and improvement will bring considerable revenue growth,and will also play an important role in the healthy development of the industry.As one of the core technologies of advertisement delivery,advertisement clickthrough rate estimation discovers the potential connection between advertisements and users based on the research of massive data,so as to improve the performance of advertisement recommendation.The increase in advertising click-through rate can directly drive the increase in subsequent conversions,thereby bringing good benefits to media platforms and advertisers,and enabling the entire business model to achieve a healthy operation.Advertising click-through rate estimation can be regarded as a binary classification problem.There is already a wealth of researches and cases on the problem.Based on the principles of the advertising click-through rate estimation model,this article proposes a comprehensive deep learning fusion model based on specific business scenarios.The superiority of the model is verified through experiments.The specific implementation process is as follows:(1)Comparative study of existing models.In summary,each approach has its advantages and disadvantages.This article attempts to integrate the idea of the Wide&Deep framework and proposes the Attention Product Factorization Machines(APFM)model.By assigning different weights to different features,both low-level and high-level features are adequately learned,which greatly improves the expressive power and scalability of the new model.(2)To avoid the influence of festivals,10 days of sample data were selected,and the training set and the test set were divided accordingly.Based on business needs and general experience,the characteristics required by the experiment are selected,and the category characteristics and numerical characteristics are encoded differently.(3)Experiments are designed and analyzed to compare seven different models,including two machine learning models and 5 deep learning models.It can be seen that the innovative APFM model performs better than others.Finally,the setting of parameters is analyzed and optimized,so as to improve the expression of the model.(4)Finally,a summary is added and the future work is presented.