摘要
为了提升电力营销系统问答机器人的智能化水平,文中基于知识图谱技术对相关智能数据分析的方法进行了研究。通过分析隐含语义在知识图谱中的稠密化向量表示方法后,针对传统方法在复杂图谱下多实体间映射关系不准确的问题,设计了一种改进的三分支并行神经网络(TBPNN)。该网络针对三元组中的头实体、尾实体及约束关系建立了结构相同的三个神经网络,且每个神经网络均包含交互层、非线性层和输出层。为了验证该网络在营销知识图谱上的数据分析效果,使用部分人工构建的知识图谱进行了仿真实验。结果表明,相较于传统的TransE算法,TBPNN网络在MeanRank上降低了39.9%,在Hit@10指标上则提升了41.5%。而在一对一、一对多与多对多的三元组分类实验上,精度分别提升了3.3%、39.0%及54.7%。
In order to improve the intelligent level of question answering robot in power marketing system,this paper studies the relevant intelligent data analysis methods based on knowledge atlas technology.After analyzing the dense vector representation method of implicit semantics in knowledge map,an improved Three Branch Parallel Neural Network(TBPNN)is designed to solve the inaccurate mapping relationship between multiple entities in complex map.The network establishes three neural networks with the same structure for the head entity,tail entity and constraint relationship in triplet.Each neural network contains interaction layer nonlinear layer and output layer.In order to verify the data analysis effect of the network on the marketing knowledge map,this paper uses some artificially constructed knowledge maps to carry out simulation experiments.The results show that compared with the traditional TransE algorithm,TBPNN network reduces 39.4%on MeanRank Hit@10 The index increased by 41.5%.In the one-to-one,one to many and many to many triple classification experiments,the accuracy is improved by 3.3%,39.0%and 54.7%respectively.
作者
薛晓茹
徐道磊
路宇
唐轶轩
XUE Xiaoru;XU Daolei;LU Yu;TANG Yixuan(Information and Communication Branch,State Grid Anhui Electric Power Co.,Ltd.,Hefei 230000,China)
出处
《电子设计工程》
2023年第22期139-143,共5页
Electronic Design Engineering
基金
国网公司科技项目(JL71-15-042)。
