Design And Implementation Of Advertising Delivery System Based On Integrated Learning

The advertising business of Internet businesses has always been one of its important revenue businesses,and ad click-through rate(CTR)is often an important benchmark for advertising platforms in decision making.Whether it is for Internet commercial platforms or advertisers,in order to maximize the benefits,it is necessary to place ads as much as possible on exposure opportunities that generate a higher frequency of clicks.For users,too many irrelevant and cluttered ads will be great,the decrease in experience has led to a surge of dissatisfaction.Therefore,accurately predicting the click-through rate of advertisements will help the enterprise platform to accurately deliver advertisements,increase the opportunities for advertisers to obtain advertising revenue,and improve user experience.Therefore,the CTR estimation model has attracted more and more attention from merchants and researchers.Logistic regression model is used as the basic model of click-through rate estimation,and advertising business data often has the problems of large amount of data,high dimensionality,and difficulty in mining the correlation between features.How to extract effective features from the complex and huge advertising data set and select efficient and accurate models to improve the click-through rate prediction is a problem that many businesses and researchers tend to ignore.Based on this background,this research aims at the problem of CTR estimation business in the advertising process,based on the prediction of the commonly used Logistic regression model,proposes an integrated learning model based on XGBoost and LightGBM,and designs and implements a prototype advertising system.The specific research content is as follows:(1)This paper uses the real advertising data set of a well-known Internet information technology company as the basic data set,processes the data from a global perspective,and displays the analysis in the form of graphs.Missing values and deviations in the processing process are cleaned;(2)This paper is an in-depth extraction of the original data set,carries out feature engineering from the three perspectives of advertising,user,and user-ad association,and expands the original 33 features to 156 dimensions.In order to prevent the over-fitting phenomenon from too many features,this paper uses the LightGBM algorithm to evaluate and filter the importance of features;(3)This paper uses LightGBM and XGBoost two classic classification models to construct a single model of advertising clickthrough rate prediction,and use Stacking to perform Model fusion.After experimenting and comparing logloss values,the fusion result is found to be a great improvement over the predictive effect of a single classifier;(4)Based on the advertising click-through rate prediction model,this paper designs and implements a set of advertising for the field of advertising CTR Delivery system.The main function of the system is to predict the probability of an advertisement being clicked by the user,divide the advertisement library,carry out personalized advertisement placement and system management,etc.The purpose of this article is to try to develop a combination of machine learning and commercial advertising systems,and apply machine learning technology to commercial advertising systems,so as to assist corporate advertising platforms in advertising CTR prediction,and basically realize that advertisers can obtain marketing revenue and advertising.A win-win situation where the platform gets higher advertising revenue and users get personalized recommendation services.The experimental results show that the fusion CTR prediction model has a higher accuracy rate,which improves the commercial value and prediction accuracy in the advertising point placement system.