基于LASSO变量选择方法的网络广告点击率预测模型研究  

Research on Search Engine Advertisement Click Rate Predicition Model Based on LASSO Variable Selection Method

作　　者：李春红[1] 吴英[1] 覃朝勇[1] 

机构地区：[1]广西大学数学与信息科学学院,广西南宁530004

出　　处：《数理统计与管理》2016年第5期803-809,共7页Journal of Applied Statistics and Management

基　　金：国家自然科学基金(71462002);广西自然科学基金(2013GXNSFAA019340)

摘　　要：与传统的的媒体营销模式相比,搜索引擎广告因其精准和投入低等特点获得巨大成功。但已有的搜索引擎广告点击率模型不能有效解决数据量大及特征维度高的问题,使预测结果的准确性大打折扣。本文构建了一种基于LASSO变量选择方法的广告点击率预测模型,能有效克服现有广告点击率模型在处理数据高维性和稀疏性方面的不足。利用某公司的竞价数据对模型进行验证,结果表明影响广告点击率的关键因素是广告关键词中的商标信息、地域信息和每点击成本。该研究结果为企业制定搜索引擎广告营销策略提供一定的理论依据。Compared with the traditional media marketing, search engine advertising achieve success because of its precision and low investment.However, the existing search engine advertising click rate prediction model is not good to handle the amount data and the data＇s high dimension makes the prediction results inaccuracy. This article contracts a model based on LASSO variable selection method can overcome the existing model＇s disadvantages on data high-dimensional and sparsity. Using a company actual biding data to verify the model, the results show that the brand and geographical information inkeywords and the cost per click have the positive effect on click rate. The results of this study provide a theoretical for enterprises to make marketing strategies.

关 键 词：点击率 分层贝叶斯 LASSO 变量选择 

分 类 号：F222.3[经济管理—国民经济] O212[理学—概率论与数理统计]