Research On Estimation Algorithm Of Click-through Rate And Conversion Rate In Computation Advertisements

Nowadays,the development of Internet companies is getting faster and faster,and a large number of technical means are emerging.The online advertising business,as primary source of income for most Internet companies,has formed a personalized and product-oriented technology delivery model in a short time.Online advertising system provides advertisers with a new marketing method and a channel to reach the target audience,which not only benefits the brand promotion of advertisers,bu t also brings great economic value to Internet companies.Ad click-through rate and conversion rate estimates are important modules of the online ad delivery system,with a large number of applications on portals,online video,and e-commerce sites.The increase in click-through rate and conversion rate estimates task can improve the marketing effectiveness of the advertising platform and give users a better experience.Mainstream way of predictive click-through rate and conversion rate composed of traditional machine learning algorithms such as logistic regression and factorization machines.These methods have been widely used in the industry,but there is still much space for improvement.With the wide application of deep learning in various fields,the depth model has also been introduced into the field of computational advertising to estimate the click-through rate and conversion rate.This paper mainly works as follows:(1)Firstly,for the problem of ad click-through rate,the depth factorization machine that introduces the attention mechanism and the pairwise loss function proposed to calculate the ad click-through rate.This method working better when hand multi-category features in real data and providing new optimization goals for ad click-through rate estimation tasks.The model was tested on the real data set published by the Tencent Advertising Algorithm Contest,The AUC of the ad click rate prediction task reached 0.766 without feature engineering,which is better than the existing deep factorization machine and traditional machine learning algorithm.(2)Then,based on the problem of advertising conversion rate estimation,this paper proposes a entir spatial multi-task learning algorithm that introduces a neural factorization machine.The algorithm combines the following advantages : modeling in a complete sample space,using a shared feature representation mechanism,and powerful data fitting ability.Experiments were conducted on Taobao's public dataset.The advertising conversion rate estimated AUC reached 0.683.The algorithm is better than the most advanced algorithms in predicting the advertising conversion rate.