互联网广告点击率预估模型中特征提取方法的研究与实现

发布时间：2018-08-22 08:20

【摘要】：互联网广告是一个具有上千亿元规模的市场,广告的点击率(CTR)是互联网广告投放效果的重要指标。在广告点击率预估模型中,特征提取是关键因素,特征的好坏直接影响到最终模型的效果。针对如何提高广告点击率预估效率问题,在Hadoop大数据平台环境中,提出了基于梯度提升决策树(gradient boost decision tree,GBDT)模型的多维特征提取方法。该方法利用原始数据构建多维基础特征库,并将基础特征库中除ID类特征以外的其余特征输入GBDT模型进行特征刷选,得到高层特征,进一步进行分类。该方法的使用不仅减少了特征提取的人工成本和时间成本,也在很大程度上提升了模型的精度。
[Abstract]:Internet advertising is a market with a scale of hundreds of billions of yuan. The click rate of advertising (CTR) is an important indicator of the effect of Internet advertising. Feature extraction is the key factor in the prediction model of ad click rate, and the quality of feature directly affects the effect of the final model. Aiming at how to improve the efficiency of ad click rate estimation, a multi-dimensional feature extraction method based on gradient elevation decision tree (gradient boost decision tree-GBDT) model is proposed in the environment of Hadoop big data. In this method, the multi-dimensional basic feature database is constructed from the original data, and the other features in the basic feature base except ID class feature are input into the GBDT model for feature selection, and the high-level features are obtained for further classification. The use of this method not only reduces the labor and time cost of feature extraction, but also improves the accuracy of the model to a great extent.
【作者单位】：北京工商大学计算机与信息工程学院;
【基金】：北京市自然科学基金重点项目B类(KZ201410011014);北京市自然科学基金青年项目(9164025) 2015年研究生科研能力提升计划资助项目国家教育部人文社会科学研究青年基金资助项目(15YJCZH224)
【分类号】：TP311.13

【参考文献】