基于门控循环单元神经网络的广告点击率预估  

Advertisement click-through rate predicting based on gated recurrent unit neural networks

作　　者：陈巧红[1] 董雯 孙麒[1] 贾宇波[1] CHEN Qiaohong;DONG Wen;SUN Qi;JIA Yubo(School of Information Science and Technology,Zhejiang Sci Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息学院,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2018年第5期587-592,共6页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：浙江省自然科学基金项目(LY17E050028)

摘　　要：为提高在线广告的投放效果,改善用户广告体验度,增加广告收益,提出了一种基于门控循环单元神经网络模型的广告点击率预估方法。该方法结合了门控循环单元网络特有的门控单元结构和广告数据时序性特点,利用按时间反向传播算法训练网络模型;提出一种门控循环单元神经网络训练步长改进算法,使得训练时间更少,模型更加精确。实验表明,与逻辑斯特回归、随机森林、朴素贝叶斯和循环神经网络模型相比,提出的方法在广告点击率预估的概率上更准确,有助于广告主、媒体和目标受众用户三方博弈,实现共赢。In order to improve the putting effect of online advertising, to improve the user experience of advertising, and to increase the revenue of advertising, an advertisement click through rate predicting method based on the gated recurrent unit neural networks is proposed in this paper. This method combines proper gate unit structure of gated recurrent unit network and time sequence characteristics of advertising data, and utilizes time based back propagation algorithm to train the network model. This paper proposes step size improvement algorithm of gated recurrent unit neural networks so that the training time is less and the model is more accurate. The experiment shows that compared with logistic regression, random forest, naive hayes and recurrent neural network models, the method proposed in this paper has more accurate advertisement click through rate prediction and contributes to three side game of advertisers, media and target audience so as to achieve win win.

关 键 词：在线广告 门控循环单元 点击率 按时间反向传播 三方博弈 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]