摘要
为有效提升搜索广告的点击率预测效果,提出一种基于特征降维和深度置信网络的模型(KTDDBN)。针对传统方法还停留在探索广告特征间的线性关系的局限性,提出使用深度置信网络寻找广告特征间更加复杂的深层关联;提取广告特征后,采用K-means聚类以及张量分解对高维特征进行降维,利用深度置信网络挖掘高阶的特征组合,提高预测模型的效果。实验结果表明,该模型在一定程度上提升了广告点击率的预测效果。
To improve the click-through rate prediction result in search advertising,a model based on feature dimension reduction and deep belief network(KTDDBN)was proposed.To solve the limitation that traditional methods still explore the linear relationship of the advertising features,deep belief network was proposed to search the complex deep associations of advertising features.After extracting advertisement features,K-means clustering and tensor decomposition were used to reduce the dimensionality of high-dimensional features and deep belief network was used to excavate high-level feature combinations,improving the effectiveness of the prediction model.Experimental results show that the proposed model can effectively improve the accuracy of CTR prediction to a certain degree.
引文
[1]ZHOU Aoying,ZHOU Minqi,GONG Xueqing.Computational advertising:A data-centric comprehensive Web application[J].Chinese Journal of Computers,2011,34(10):1805-1819(in Chinese).[周傲英,周敏奇,宫学庆.计算广告:以数据为核心的Web综合应用[J].计算机学报,2011,34(10):1805-1819.]
[2]ZHU Zhibei,LI Bin,LIU Xuejun,et al.Resarch on clickthrough rate prediction of Internet advertising based on LDA[J].Application Research of Computers,2016,33(4):979-982(in Chinese).[朱志北,李斌,刘学军,等.基于LDA的互联网广告点击率预测研究[J].计算机应用研究,2016,33(4):979-982.]
[3]PAN Shumin,YAN Na,XIE Jinkui.Study on advertising click-through rate prediction based on user similarity and feature differentiation[J].Computer Science,2017,44(2):283-289(in Chinese).[潘书敏,颜娜,谢瑾奎.基于用户相似度和特征分化的广告点击率预测研究[J].计算机科学,2017,44(2):283-289.]
[4]Jahrer M,Toscher A,Lee J Y,et al.Ensemble of collaborative filtering and feature engineered models for click through rate predition[C]//Proceedings of the 18th ACM SIGKDD Conference on Knowledge Discovery and Data Mining,KDD Cup Workshop,2012.
[5]YUE Kun, WANG Chaolu,ZHU Yunlei,et al. Clickthrough rate prediction of online advertisements based on probabilistic graphical model[J].Journal of East China Normal University(Natural Sciense),2013(3):15-25(in Chinese).[岳昆,王朝禄,朱运磊,等.基于概率图模型的互联网广告点击率预测[J].华东师范大学学报(自然科学版),2013(3):15-25.]
[6]Rendle S.Social network and click-through prediction with factorization machines[EB/OL].[2012-08-08].https://kaggle2.blob.core.windows.net/competitions/kddcup2012/2748/media/Rendle.pdf.
[7]Trofimov I,Kornetova A,Topinskiy V.Using boosted trees for click-through rate prediction for sponsored search[C]//International Workshop on Data Mining for Online Advertising and Internet Economy.US:ACM,2012.
[8]Lee K C,Orten B,Dasdan A,et al.Estimating conversion rate in display advertising from past erformance data[C]//Acm Sigkdd International Conference on Knowledge Discovery&Data Mining.US:ACM,2012:768-776.
[9]XU Qingyong,JIANG Shunliang, HUANG Wei,et al.Image classification algorithm for deep belief network based on multifeature fusion[J].Computer Engineering,2015,41(11):245-252(in Chinese).[许庆勇,江顺亮,黄伟,等.基于多特征融合的深度置信网络图像分类算法[J].计算机工程,2015,41(11):245-252.]
[10]Williams C,Agakov F.An analysis of contrastive divergence learning in gaussian boltzmann machines[J].University of Edinburgh,2002.
[11]Carreira-Perpinan M A,Hinton G E.On contrastive divergence learning[J].Artificial Intelligence&Statistics,2005.
[12]Shen S,Hu B,Chen W,et al.Personalized click model through collaborative filtering[C]//ACM International Conference on Web Search and Data Mining.US:ACM,2012:323-332.