摘要
因子分解机(Factorization Machines,FM)在不同的输入实例中对每个特征产生单一的固定表示,忽略了特征的多语义特性,限制了点击率(Click-through Rate,CTR)预估模型的表示和预测能力.针对这一问题,提出一种深度协同感知的因子分解机模型,引入多语义交互感知网络和三重输入感知网络,通过多语义的特征域交互并融合不同层级的特征交互信息学习不同样本的感知因子,从而获得更加准确的特征表示.通过模型对比实验和消融实验表明:该模型可以有效提升点击率预测的准确性.
Factorization Machines(FM)produce a single fixed representation of each feature in different input instances,ignoring the multi-semantic nature of features and limiting the representation and predic-tion ability of Click-through Rate(CTR)estimation models.To solve this problem,a factor decomposi-tion machine model of deep collaborative perception is proposed,which introduces multi-semantic interac-tion perception network and triple input perception network,and helps to learn the perception factors of different samples through multi-semantic feature domain interaction and fusion of feature interaction infor-mation at different levels,so as to obtain more accurate feature representation.Through model comparison experiment and ablation experiment,this model can effectively improve the accuracy of click-through rate prediction.
作者
李春秋
卜天然
何军
LI Chunqiu;BU Tianran;HE Jun(Anhui Business College of Vocational Technology,Wuhu 241002,China)
出处
《通化师范学院学报》
2024年第2期82-90,共9页
Journal of Tonghua Normal University
基金
安徽省教育厅科学研究重点项目(2022AH052739,KJ2020A1082)
安徽省高校优秀青年人才支持计划重点项目(gxyqZD2018130)
安徽省职业与成人教育学会一般项目(Azcj2022129)。
关键词
因子分解机
样本输入感知
多语义交互感知
协同感知
Factorization Machines
input-aware
multi-semantic interactive awareness
cooperation-aware
