Application Research Of Search Advertisement Conversion Rate Prediction Based On Integrated Learning Algorithm

In today’s information-rich Internet environment,the scarcest resource is the attention of users(consumers),that is,traffic.Search advertising is a mechanism for advertisers(merchants)to pay attention to the user’s attention,allowing advertisers’ ads to reach consumers.The conversion rate(CVR)of a search ad is the probability that an ad product will be purchased after being clicked by the user.Accurately predicting conversion rates,on the one hand,enables advertisers to match the users who are most likely to purchase their own products,thereby increasing the advertiser’s input-output ratio(ROI);on the other hand,it also allows users to quickly find the strongest willingness to buy.Products,thereby enhancing the user’s shopping experience in the e-commerce platform.How to make better use of massive transaction data to efficiently and accurately predict the user’s purchase intention has important guiding significance for effectively improving the advertiser’s ROI and the user’s shopping experience.Based on this direction,based on the domestic and foreign literature research,through the massive real transaction data of Taobao platform in 2018,the data mining technology is used to construct the search advertisement conversion rate prediction model based on the integrated learning algorithm to estimate the user’s purchase intention.This paper mainly includes Feature engineering and model building two parts.The feature engineering part mainly includes the following two aspects.On the one hand,data exploratory analysis and data preprocessing.Visualize the distribution and comparison of the dataset,and conduct exploratory analysis of the structure and regularity of the sample dataset,and provide support for the subsequent selection of appropriate data preprocessing methods and in-depth business mining of the search advertising conversion rate prediction model.In data preprocessing,data processingfor different feature variables is performed on data deletion,discretization,and standardization.On the other hand,feature extraction and selection.In order to enrich the characteristics of the original dataset,basic features,statistical features and combined features are extracted from the original features such as advertising product information,user information,context information and store information.The model building part,based on the previous feature engineering,uses the finalized feature variables to use the random forest model,XGBoost model,LightGBM model and Stacking fusion model for parameter tuning and sample training to obtain the corresponding optimal model to convert the search advertisement.Rate to make predictions.According to the comparative analysis of predictive ability indicators,LightGBM has the best prediction effect in three single models,with logloss reaching 0.08486 and AUC value reaching 0.7088379.After the integration of the Stacking model,the prediction ability of the single model LightGBM is significantly improved compared with the forecasting ability.The logloss is reduced by 0.081% and the AUC value is increased by 2.088%.The application research on the search conversion rate prediction has certain expansion significance.Finally,from the importance of feature engineering,the comparative study of the search advertising conversion rate prediction model and the in-depth business background mining,the paper summarizes the full text,achieves the purpose of deepening the user’s purchase intention and proposes feasible suggestions for improving the advertiser ROI.The shortcomings suggest corresponding follow-up research.