摘要
在智慧数据库系统预测任务中,查询的向量表示往往需要多个领域的特征共同作用,才能得到较好的效果。针对多领域特征的表征与综合问题,本文通过深度神经网络的理论与技术,为语义、结构等领域特征设计了张量化的表示方案,并提出权基综合、感知机综合、虚线综合3种多领域特征综合方案。大量真实数据的实验结果表明,所提出的多领域特征与综合方法,能够有效地提取与转化多领域的查询相关特征,具有较好的收敛效果,能够为智慧数据库系统的其它预测任务提供查询向量表示方面的支持与嵌入。
In the prediction task of a smart database system,the vector representation of a query often requires a combination of features from multiple fields to achieve better results. Aiming at the problem of representation and synthesis of multi-domain features,this paper uses the theory and technology of deep neural networks to design a quantitative representation scheme for domain features such as semantics and structure,and proposes three kinds of multi-domain synthesis,i. e. weight-based synthesis,perceptron synthesis,and jump line synthesis. Experimental results on a large amount of real data show that the proposed multidomain feature representation and synthesis method can effectively extract and transform multi-domain query-related features which has a good convergence effect and can provide query vectors for other prediction tasks of smart database systems.
作者
张翔熙
王宏志
ZHANG Xiangxi;WANG Hongzhi(Massive Data Computing Research Center,Harbin Institute of Technology,Harbin 150001,China)
出处
《智能计算机与应用》
2021年第9期174-176,183,共4页
Intelligent Computer and Applications
基金
CCF-华为数据库创新研究计划项目(DBIR2019005B)。
关键词
智慧数据库
深度神经网络
多模态
特征综合
smart database system
deep neural network
multimodal
feature synthesis
