基于特征学习的广告点击率预估技术研究  

Research on Advertising Click-Through Rate Estimation Based on Feature Learning

作　　者：张志强[1] 周永[1] 谢晓芹[1] 潘海为[1] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,哈尔滨150001

出　　处：《计算机学报》2016年第4期780-794,共15页Chinese Journal of Computers

基　　金：国家自然科学基金(61370084;61202090;61272184);教育部新世纪人才支持计划(NCET-11-0829);黑龙江省自然科学基金(F201130);中央高校基本科研业务费专项资金(HEUCF100609;HEUCFT1202)资助

摘　　要：搜索广告中的点击率预估问题在信息检索和机器学习等领域一直是研究的热点.目前通过设计特征提取方案获得特征和针对用户点击行为建模等方法,并没有充分考虑广告数据具有的高维稀疏性、特征之间存在高度非线性关联的特点,致使信息利用不充分.为了降低数据稀疏性和充分挖掘广告数据中隐藏的规律,该文提出了面向广告数据的稀疏特征学习方法.该方法基于张量分解实现特征降维,并充分利用深度学习技术刻画数据中的非线性关联,以解决高维稀疏广告数据的特征学习问题,实验结果验证了文中提出的方法能够有效地提升广告点击率的预估精度,达到了预期效果.The issue of click through rate estimation in sponsored search has been widely studied in information retrieval,machine learning and query recommendation etc.Some related studies,such as the methods in which features are obtained by setting the feature extraction scheme or aiming at user behavior modeling,did not take into account those essential characteristics including the sparseness of advertising data and highly nonlinear association between features.In order to fully mining the hidden rules in advertising data,this paper proposes a method that can learn the sparse feature of advertising data.Our method combines dimension reduction based on tensor decomposition and takes full advantage of feature learning to portraying the nonlinear associated relationship of data to solve sparse feature learning problems.Finally,the comparison experiment shows this method has the desired effect of improving the accuracy of CTR estimation.

关 键 词：搜索广告 点击率 张量分解 深度学习 社交网络 社会媒体 计算广告学 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]