摘要
点击率预测(Click-Through Rate,CTR)是在线展示广告中的一项关键任务,CTR预测任务中涉及的数据通常有多个特征,对其中的重要特征提取、建模的方式极大地影响了CTR预测的准确性.以往方法在特征重要性提取过程中存在信息干扰问题.针对这一问题,提出了一种特征重要性动态提取的广告点击率预测模型.该模型将门控机制引入CTR模型对特征重要性进行初步筛选,同时利用了挤压提取网络获取特征重要性,并通过双线性交互获得重要性特征之间的关联信息,最后使用了隐藏门控残差网络学习高阶信息交互.通过对两个真实广告数据集进行的广泛实验,证明了其能够比传统的点击率预测模型以及最新的基于深度学习的预测模型获得更好的准确度.
Click-through rate prediction(CTR)is a crucial task in online display advertising.The data involved in CTR prediction usually has multiple features,and the method of extracting and modeling important features greatly affects the accuracy of CTR prediction.The previous methods have the problem of information interference in the process of feature importance extraction.In order to solve this problem,a novel advertising click-through rate prediction model based on dynamic extraction of feature importance is proposed.The model introduces the gating mechanism into the CTR model to initially screen the importance of features.At the same time,it uses a squeezing extraction network to obtain the importance of features,and obtains the association information between the important features through bilinear interaction,and finally uses the hidden gating residual network learns high-level information interaction.Through extensive experiments on two real advertising data sets,it is proved that it can obtain better accuracy than the traditional click-through rate prediction model and the state-of-the-art prediction model based on deep learning.
作者
蒋兴渝
黄贤英
陈雨晶
徐福
JIANG Xing-yu;HUANG Xian-ying;CHEN Yu-jing;XU Fu(School of Computer Science&Engineering,Chongqing University of Technology,Chongqing 400054,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2022年第5期976-984,共9页
Journal of Chinese Computer Systems
基金
国家社会科学基金项目(17XXW005)资助
重庆理工大学研究生创新项目(clgycx20203118)资助。
关键词
点击率预测
特征重要性提取
门控机制
残差网络
深度神经网络
click-through rate prediction
feature importance extraction
gating mechanism
residual network
deep neural network
作者简介
蒋兴渝,男,1994年生,硕士研究生,CCF会员,研究方向为推荐系统、点击率预测,E-mail:jiangcmd@qq.com;黄贤英,女,1967年生,硕士,教授,CCF高级会员,研究方向为复杂网络、舆情传播;陈雨晶,女,1995年生,研究方向为深度学习、股票金融预测;徐福,男,1996年生,硕士研究生,研究方向为情感分类、点击率预测.
