基于店铺特征和用户需求的广告转化率预测  

Prediction of Advertising Conversion Rate Based on Store Characteristics and User Needs

作　　者：孙玥 杨国为[1,2] 何鎏一 SUN Yue;YANG Guowei;HE Liuyi(School of Electronic Information,Qingdao University,Qingdao 266071,China;School of Information Engineering,Nanjing Audit University,Nanjing 210000,China)

机构地区：[1]青岛大学电子信息学院,山东青岛266071 [2]南京审计大学信息工程学院,江苏南京210000

出　　处：《青岛大学学报（工程技术版）》2019年第3期16-20,共5页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：国家重点研发计划项目(2017YFC080-4000);国家自然科学基金面上项目(61772277);江苏省基础研究计划项目(BK20171494)

摘　　要：针对现有搜索广告转化率预测模型和分类模型未考虑店铺特征和用户需求,为了更好的预测广告的转化率,本文基于店铺特征和用户需求对广告转化率进行预测。以阿里搜索广告为研究对象,提出基于店铺特征和用户需求的数据预分析的特征处理方式,对特征进行预分析,即对用户和店铺的相关特征进行初次预测处理,分别求出转化率,以此作为新特征。XGBoost算法泛化性能高,损失函数同时用到一阶导和二阶导,可以加快优化速度,所以运用该算法构建基于店铺特征和用户需求的阿里搜索广告转化率预测模型和转化率分类模型。通过对比预测结果在对数似然损失(Logarith mic loss,Logless)的指标,该预测模型的正确预测率和正确分类率显著提升。本文使用的特征处理方式能够充分挖掘商品信息,能够更好的实现广告转化率的预测,有利于提高广告的竞争力。In view of the fact that the existing search advertising conversion prediction model and classification model do not consider the store characteristics and user needs,in order to better predict the advertising conversion rate,this paper forecasts the advertising conversion rate based on the store characteristics and user needs.Taking Ali search advertisement as the research object,this paper proposes the feature processing method based on the pre-analysis of the characteristics of the store and the user′s needs,and pre-analyzes the feature,that is,the initial prediction process is performed on the relevant features of the user and the store,and the conversion rate is determined separately.This is a new feature.The XGBoost algorithm has high generalization performance,and the loss function uses both the first-order and second-order derivatives to speed up the optimization.Therefore,the algorithm is used to construct the Ali search ad conversion rate prediction model and the conversion rate classification model based on store characteristics and user requirements.By comparing the prediction results with the Logless index,the correct prediction rate and the correct classification rate of the prediction model are significantly improved.The feature processing method used in this paper can fully exploit the product information,and can better predict the conversion rate of the advertisement,which is conducive to improving the competitiveness of the advertisement.

关 键 词：搜索广告 预分析 XGBoost 转化率 店铺特征 用户需求 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]