纵向数据模型的变量选择及其在网络营销中的应用
发布时间：2018-07-28 16:37
【摘要】：在互联网大数据的背景下,纵向数据因其能将截面数据和时间序列数据有效的结合起来,所以在网络营销中占有非常重要的地位.特别在网络营销中也经常会由于高维数据的稀疏性,导致高维空间中的数据处理方法与低维空间中存在显著差异,有必要进一步解决和研究.传统技术在大数据环境下并不能很好地对高维数据进行研究,因此本文将新型的变量选择方法与传统的纵向数据结合,具体的研究内容和结果如下:首先,将Elastic Net方法应用于网络营销中经常出现的纵向数据,不仅能够更好地理解大数据对各种营销活动的影响,同时也让企业更好地发挥其有效性.证明了平衡纵向数据模型的Elastic Net估计具有组效应性质.数据模拟验证了 Elastic Net方法能将强相关变量全部选入纵向数据模型,而Lasso方法无此功效.其次,尽管Elastic Net方法相对于Lasso方法在处理强相关变量组问题时具有显著的优越性,但二者均不具备Oracle性质.本文将Adaptive Elastic Net方法与纵向数据模型进行结合,证明了 Adaptive Elastic Net方法能更有效的处理强相关变量和零变量,即具有组效应性质及Oracle性质.并通过数值模拟,表明了 Adaptive Elastic Net方法与纵向数据模型结合后的估计值明显优于Lasso方法和Elastic Net方法.最后,我们利用网络营销中广告点击率的实际例子,将纵向数据与广告点击率相结合,利用Elastic Net方法进行变量选择,筛选出重要的关键词,从而更好的提高广告点击率.表明Elastic Net方法用于纵向数据模型是可行的,且模型的拟合效果和预测能力均强于传统的纵向数据模型.同时它还实现了 Elastic Net方法在网络营销中的应用.
[Abstract]:Under the background of Internet big data, vertical data play a very important role in network marketing because it can effectively combine cross-section data and time series data. Especially in network marketing, because of the sparsity of high-dimensional data, there are significant differences between the methods of data processing in high-dimensional space and low-dimensional space, so it is necessary to further solve and study. The traditional technology can not study the high-dimensional data well under the big data environment, so this paper combines the new variable selection method with the traditional longitudinal data. The specific research contents and results are as follows: first, Applying the Elastic Net method to the vertical data which often appears in the network marketing can not only better understand the influence of big data on various marketing activities, but also enable enterprises to give full play to its effectiveness. It is proved that the Elastic Net estimation of the balanced longitudinal data model has the property of group effect. The data simulation verifies that the Elastic Net method can select all the strongly correlated variables into the longitudinal data model, but the Lasso method does not. Secondly, although the Elastic Net method is superior to the Lasso method in dealing with the problem of strongly correlated variable sets, neither of them has the Oracle property. In this paper, the Adaptive Elastic Net method is combined with the longitudinal data model, and it is proved that the Adaptive Elastic Net method can deal with strong correlation variables and zero variables more effectively, that is, it has the property of group effect and Oracle property. Numerical simulation shows that the estimated value of Adaptive Elastic Net method combined with longitudinal data model is obviously better than that of Lasso method and Elastic Net method. Finally, we make use of the actual example of advertising click rate in network marketing, combine longitudinal data with ad click rate, use Elastic Net method to select variables, screen out important keywords, so as to improve the ad click rate better. It is shown that the Elastic Net method is feasible for the longitudinal data model, and the fitting effect and prediction ability of the model are better than that of the traditional longitudinal data model. At the same time, it also realizes the application of Elastic Net method in network marketing.
【学位授予单位】：广西大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;F724.6;F224
[Abstract]:Under the background of Internet big data, vertical data play a very important role in network marketing because it can effectively combine cross-section data and time series data. Especially in network marketing, because of the sparsity of high-dimensional data, there are significant differences between the methods of data processing in high-dimensional space and low-dimensional space, so it is necessary to further solve and study. The traditional technology can not study the high-dimensional data well under the big data environment, so this paper combines the new variable selection method with the traditional longitudinal data. The specific research contents and results are as follows: first, Applying the Elastic Net method to the vertical data which often appears in the network marketing can not only better understand the influence of big data on various marketing activities, but also enable enterprises to give full play to its effectiveness. It is proved that the Elastic Net estimation of the balanced longitudinal data model has the property of group effect. The data simulation verifies that the Elastic Net method can select all the strongly correlated variables into the longitudinal data model, but the Lasso method does not. Secondly, although the Elastic Net method is superior to the Lasso method in dealing with the problem of strongly correlated variable sets, neither of them has the Oracle property. In this paper, the Adaptive Elastic Net method is combined with the longitudinal data model, and it is proved that the Adaptive Elastic Net method can deal with strong correlation variables and zero variables more effectively, that is, it has the property of group effect and Oracle property. Numerical simulation shows that the estimated value of Adaptive Elastic Net method combined with longitudinal data model is obviously better than that of Lasso method and Elastic Net method. Finally, we make use of the actual example of advertising click rate in network marketing, combine longitudinal data with ad click rate, use Elastic Net method to select variables, screen out important keywords, so as to improve the ad click rate better. It is shown that the Elastic Net method is feasible for the longitudinal data model, and the fitting effect and prediction ability of the model are better than that of the traditional longitudinal data model. At the same time, it also realizes the application of Elastic Net method in network marketing.
【学位授予单位】：广西大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;F724.6;F224
【参考文献】
相关期刊论文 前10条
1 李春红;吴英;覃朝勇;;基于LASSO变量选择方法的网络广告点击率预测模型研究[J];数理统计与管理;2016年05期
2 许亮亮;;搜索引擎营销应用分析——以Burberry公司为例[J];传播与版权;2016年09期
3 许丽庆;田凯;熊琴;;Elastic Net方法在纵向数据模型中的应用[J];数学理论与应用;2016年02期
4 王s，
本文编号：2150905
本文链接：https://www.wllwen.com/jingjilunwen/guojimaoyilunwen/2150905.html
