Research On Ad Click-through Rate Prediction And Bidding Algorithm In Real-time Bidding

With the development of programmatic advertising and the rapid rise of real-time bidding advertising,the online advertising market is moving forward with data driven and algorithmic direction.The real time bidding mode provides the demand side with the real time and bidding way to buy the advertisement display,and realizes the accurate and personalized ad serving.In order to optimize the performance of real-time bidding,the demand side platform needs exquisite big data technology,meticulous advertising management capabilities,solid mathematical modeling skills and excellent intelligent bidding algorithm.The real-time bidding algorithm is one of the core product strategies of the demand side platform.It directly reflects the demand side platform's ability to use various technologies,and determines the revenue gained by advertisers,which is the key for the demand side platform to maintain market competitiveness.The ad clickthrough rate or CTR is an important factor in the design of real time bid bidding algorithm.Therefore,accurate CTR prediction technology is also critical.There are two major challenges for estimating CTR in real-time bidding advertising.First,the real-time bidding environment has high time efficiency requirements for estimating the click rate based on a single advertisement display,the model should have the ability to quickly adjust in real time;second,imbalanced online data stream are common in real-time bidding advertising,which is very strict with the algorithm's performance on the minority samples.Based on the above two difficulties,this paper focus on the method of predicting CTR,and provides an effective solution.First,in order to enable the predictive model to train efficiently large scale data sets in real time,an online learning algorithm is used to search the optimal solution of the parameter space.Secondly,the UOB learning strategy is proposed to mitigate the impact of the imbalanced online data streams on the predictive performance.Finally,logistic regression model is trained to estimate the CTR based on UOB learning strategy and the experimental results show that the model is optimized both in terms of predictive performance and time complexity.The simple linear bidding strategy considers both budget constraints and the value of ad requests,is widely used by demanders.However,when the budget constraint is below a certain level,the parameters of the linear bid model are reduced,the calculated bid may be lower than the market price,and the bid failure rate and the loss degree of high CTR advertisement exposure are increased,the optimization of advertising effectiveness can not be achieved.This paper proposes a loss reducing bidding algorithm based on probability control in order to make up for the shortcomings of linear bidding strategy.Experiments show that the algorithm can effectively reduce the loss of clicks in ad campaign and improve advertiser's revenue.