在线广告中改进数据分层的动态点击率评估算法

Dynamical Click-through Rate Estimation Algorithm by Improving Data Hierarchy in Online Advertising

在线广告作为广告主向用户传达信息的载体,在这个信息过载的时代具有重要的意义.点击率是在线广告中的一个重要的指标,能够帮助广告主进行广告性能优化和广告预算投放.然而,由于其数据稀疏性的特征,使得点击率的评估难以达到很高的准确度.为了准确地评估点击率,本文不仅从机器学习中基于决策树分类器的角度加以改进,而且从广告数据本身的角度出发,充分考虑数据本身的层次关系,增加了对点击与展现随时间的演化建模,提出了一种动态点击率模型算法.在真实互联网广告数据中对所设计的算法进行实现,并与传统机器学习的算法做实验对比,AUC值提升幅度达到17%,验证了本文提出的算法是对数据稀疏问题的有效解决方案.

As a carder used to convey information to users by advertisers,online advertising has the vital significance in the era of in-formation overload. Click-through rate is an important indicator of online advertising, to help advertisers on advertising performance e-valuation and budget planning. However,due to data sparsity,the accuracy of click-through rate estimation is not so high. In order toestimate accurately the click-through rate, We not only improve from the perspective of decision tree classification tool in machinelearning, but also from the perspective of the data itself, considering full the hierarchies of data itself, increasing click and impression e-volution with time modeling, proposing a dynamic click-through rate estimating algorithm. We implement the algorithm in the real on-line advertising data. Comparedto traditional machine learning algorithm, the result shows that AUC value increases to 17%, it is veri-fied that the algorithm I propose is an effective solution to the data sparsity problem.