摘要
现有广告转化率预估模型缺乏对深层特征间相互作用的研究,针对这一问题提出了一种新的混合模型.通过高效的梯度提升机(light gradient boosting machine,LightGBM)模型提取高阶组合特征,并结合基于区域的因子分解机(field-aware factorization machines,FFM)模型有效处理稀疏数据的优点进行转化率的预估.为了验证模型的有效性和泛化能力,在两个数据集上讨论了参数对预估结果的影响,并将模型与其他模型进行对比实验.实验结果表明提出的混合模型的预估结果更准确.
Many existing models of predicting advertising conversion rate lack research on the interaction among deeper features. Hence, a new hybrid model was proposed for this problem. High-level combination features were extracted using a light gradient boosting machine(LightGBM) model, and combining with the advantages of field-aware factorization machines(FFM) model. It can effectively process sparse data and predict the conversion rate. In order to verify the effectiveness and generalization ability of the hybrid model, the model was tested on two data sets for discussing the influence of parameters on model prediction results and was compared with other models. The experimental results show that the hybrid model is more accurate.
作者
李雄飞
周晋男
张小利
LI Xiong-fei;ZHOU Jin-nan;ZHANG Xiao-li(College of Computer Science and Technology,Jilin University,Changchun 130012,China)
出处
《东北大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2019年第7期942-947,共6页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(61801190)
吉林省自然科学基金资助项目(20180101055JC)
吉林省优秀青年人才基金资助项目(20180520029JH)
中国博士后科学基金资助项目(2017M611323)
关键词
转化率预估
高效的梯度提升树
基于区域的因子分解机
混合模型
高阶组合特征
conversion rate prediction
light gradient boosting decision tree
field-aware factorization machines
hybrid model
high-level combination features
