Click Through Rating System Based On Distributed Logistic Regression Model

Advertising has always been an important source of income for Internet companies.Like Google,Yahoo,Baidu,Weibo,etc.,these leaders in the Internet are using advertising as the main revenue.Driven by machine learning and artificial intelligence,more and more companies are investing in research on how to improve their ability to cash out.Compared with traditional advertising,the environment in which mobile advertising users are located is more complicated,so more accurate delivery techniques are needed.Therefore,it is a huge challenge to study how to get the right advertisements from the platform's advertising database in real time.This article is based on such a complex environment,considering how to design and implement a system that can calculate an accurate click rate for each advertisement in the candidate advertisement set.The CTR system needs to make reasonable use of the user's own information and the user's browsing page information,estimate the probability of the advertisement being clicked by the user,sort the advertisements according to the calculated probability and bidding factor,and select the top N.This is the click-through-rate(CTR)estimate.Based on the estimated value of CTR,the candidate ads are ranked.From the user's point of view,it is more likely to see the advertisements that are interested to them.From the perspective of the advertisers,the advertisements achieve better communication effects.From the platform's point of view,the retention rate of platform users is high,and advertisers will be more and more involved.Platform benefits will increase.In order to achieve the maximization of these three interests,the author designed and implemented the advertisement click rate estimation system based on the distributed logistic regression model.First,introduce the related technologies and evaluation indexes of the advertisement click-rate estimation system;then,elaborate the functional and non-functional requirements analysis of the system;then,divide the system modules by the results obtained from the demand analysis.According to the results obtained in the previous stages,the detailed design of each functional module is performed,and high-quality code is written;finally,the system is tested for functionality and non-functionality.In the test,by comparing the four gradient descent algorithms for SGD,Ftrl,Adam,and Adagrad,the Adam algorithm that is most suitable for the distributed logistic regression model.And the calculation method of the logistic regression algorithm of the training module is improved.The training speed of the implemented CTR system has been significantly improved.