Research On The Application Of The Logistic Regression Model Based On Feature Optimization In Advertising Click-through Rate

With the rapid development of information technology,computer technology and Internet technology,Internet advertising has become one of the main sources of Internet revenue.Click-through Rate Prediction(CTR)represents the popularity of advertising and the number of times users watch it.As a result,most advertising revenue is paid by click,and the revenue generated by advertising media is the product of single click rate expense and click rate.For this reason,it is very important to accurately estimate the click-through rate of advertisements,which is a three-way thing.For advertisers,the product has been effectively promoted,and the potential users have been increased,which can effectively improve the revenue generation.For search engine companies,higher AD clicks mean more revenue;For users,the ads displayed according to the click rate are exactly what they need in the current situation,which improves the user's experience and makes them more willing to click.The estimation of CTR is a complex and wide-ranging problem.Logistic Regression model(Logistic Regression,LR),as a binary classification prediction model,is often used in the click rate forecast problems.compared with the traditional linear model,LR used logistic transformation function values mapped to interval,mapping the function value is the estimated CTR.However,the traditional logistic regression model is limited by the logarithmic linear relationship between CTR and the variables,and the effectiveness of the training model is often need to be improved.Aiming at this problem,this paper compares several different feature extraction,integration and optimization techniques.Using the principal component,Decision Tree,and Gradient Boost Decision Tree,we extracted the few characteristics from the complicated variables in the original data,and then incorporated with the LR model to do modeling analysis.Based on a real advertising click data,the LR model based on GBDT can effectively improve the estimation accuracy of the traditional LR model.