Advertising Click Through Rate Prediction Based On Feature Selection And Integration Algorithm

In recent years,the network marketing is becoming hot in the background of massive data.Search advertising,as a new form of advertising,has shown great commercial value and potential,and has become one of the main sources of income in the Internet industry.As the key technology of search advertising,CTR is not only related to advertising ranking,but also affects the charge of advertising click.It is very meaningful to estimate the click rate of advertisement effectively in the mass data.This paper is based on real data,and uses a variety of feature selection methods and different models to study the advertising click rate prediction model.Through comparing the evaluation indexes of the model,the best prediction model is obtained among the many alternative models,and the final prediction is carried out.The main contents of this paper are as follows.1.The traditional model gradient tree and logic regression are used to build the model,and the evaluation indexes are obtained.2.The feature selection methods are mutual information,latent method and filtering and packaging method.Nine separate CTR models are built by using three integration algorithms,namely xboost,catboost and lightgbm.3.The best three models are selected from the above 9 models to construct the hybrid model.The results show that the effect of single model is 11% higher than that of traditional prediction model,and that of mixed model is 1.15% higher than that of the single model.The conclusion of this paper can provide reference for Internet companies in CTR estimation.Mutual information,latent method,filtering and packaging method can be used in CTR estimation;In model training,xbboost,catboost and lightgbm can be used,or different models can be integrated into mixed models to improve the efficiency of CTR prediction.