The Conversion Rate Prediction Of Search Advertising

Advertising,as the name suggests,is to inform the general public of a certain thing.In a narrow sense,advertising refers to the advertisement for profit,such as commercial advertisement or economic advertisement.It is a means by which commercial enterprises promote goods or services to consumers or users through advertising media by paying fees.Commodity advertisement is right what an economic advertisement is.In the 21 st century,the Internet industry,such as big data and e-commerce,has made great progress.Search advertising has become an important part of commercial Internet advertising.It has become the most popular types of advertising when people use Internet products or looking through web pages.So the prediction of conversion rate has become one of the hottest issue within the object of scholars' research.The conversion rate and click rate are different.Since the conversion rate is directly oriented to purchase.It meets its differences from click rate.Therefore,high conversion rate is the key research object of advertisers and advertising media because of its higher output ratio that directly brings to them.This paper mainly studies the conversion rate prediction model of search advertising and its feature processing.Taobao is currently a large-scale e-commerce platform in China,searching advertising plays an important role of its promotion.Due to the fact of the company's data publicity recently,this paper chooses to use the Alibaba platform's 2018 search advertisement data as research data.To begin with,this paper makes a descriptive statistical analysis of the original data from four perspectives including1)user's information data;2)commodity information data;3)store information data and 4)supplementary commodity context data information.After that,the author processed missing data,text feature and time.Based on the data preprocessing,the data is extracted by feature processing which includes hidden information extraction such as detailed description of commodity description,segment processing such as segmentation of the store evaluation quantity level,and combined feature extraction of the click rate.The final data is: 478138 samples,10075 different advertising products,and 3959 different shops,270 explanatory variables.Secondly,in this paper,Python is used to construct the search advertising conversion rate prediction model based on LightGBM and XGBoost algorithms.Regarding the important hyperparameters,this paper also adopts the grid search algorithm to search for better parameters which ensure a better prediction model.In terms of evaluation,the paper uses logloss loss value,auc,rmse(mean square error)and training time to compare the advantages and disadvantages of LightGBM model and XGBoost model.The results show that although the logloss value,auc value and rmse mean square error value are not much different,the three evaluation indexes of the LightGBM algorithm are slightly better than the XGBoost algorithm.In particular,LightGBM is very computationally efficient and is nearly a dozen times faster than XGBoost.Both LightGBM and XGBoost are recent algorithms for machine learning,and have been widely praised by scholars from all walks of life.In this paper,LightGBM algorithm and XGBoost algorithm are applied to the prediction model of search advertising conversion rate.On the one hand,the selection range of such prediction model increases,and on the other hand,the application range of LightGBM and XGBoost algorithm is expanded.However,due to the fact that preliminary research of LightGBM and XGBoost algorithm are still in their initial phases and prediction research of search advertising conversion rate is also in the primary stage of quantitative research,there are few relevant academic literatures that can be referenced at present,insufficiency may exist in the research.This paper is an attempt of these two machine learning algorithms in the research of search advertising conversion rate prediction which does not combine it with other machine learning algorithms to study and construct.With further study on scholars' research of LightGBM and XGBoost algorithms and a deeper research on search advertising conversion rates,I believe that more optimized algorithms will be applied in the search advertising conversion rate forecasting model.