Design And Implementation Of Content Click Through Rate Prediction System Based On Logistic Regression With Elastic Net

With the advent of the information age,the amount of information is increasing exponentially.When users face such huge information,how to select effective information is an urgent problem to be solved.Recommendation system comes from it.Recommendation system helps users filter out useful information from huge information.In this paper,click through rate prediction system is the core of recommendation system.The content click through rate prediction system estimates the probability of clicking the candidate content,and sort by the probability.Then displays the selected content to the user.The content click through rate prediction system includes two parts:off-line module and online module.The off-line module mainly includes feature extraction,feature combination,model training,model updating and other functions.The online module mainly includes feature calculation,model calling,and click through rate calculation.This paper focuses on the design and implementation of content click through rate prediction system.First,we analyze the needs of the system,and research the related technologies that the implementation needs,and determine to use which algorithm.Then the overall design,detailed design and implementation of the system are carried out.Finally,the test and result analysis are carried out.In this paper,we select features by Chi-square test.In the model training,logistic regression algorithm is used to solve the problem.In order to prevent overfitting,elastic net regularization is applied to punish the loss function.When the trained model needs updating to online,we use Read-Copy Update to update the model files.We use AUC to evaluate offline models and use actual CTR to evaluate online services.At last,we get the content click through rate prediction system,which makes the content actual click through rate increase,and improves the accuracy of content recommendation.