计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (8): 145-152.DOI: 10.3778/j.issn.1002-8331.2001-0180

• 模式识别与人工智能 • 上一篇    下一篇

基于动态投影嵌入的多维异质网络可视化研究

余磊,许光銮,王洋,林道玉,李峰   

  1. 1.中国科学院 空天信息创新研究院,北京 100094
    2.中国科学院 网络信息体系技术重点实验室,北京 100190
    3.中国科学院大学 电子电气与通信工程学院,北京 100049
  • 出版日期:2021-04-15 发布日期:2021-04-23

Research on Multidimensional Visualization of Heterogeneous Network Based on Dynamic Projection Embedding

YU Lei, XU Guangluan, WANG Yang, LIN Daoyu, LI Feng   

  1. 1.Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
    2.Key Laboratory of Network Information System Technology(NIST), Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China
    3.School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2021-04-15 Published:2021-04-23

摘要:

异质网络是包含多种类型节点和边的复杂信息网络,因此异质网络的可视化通常涉及异质信息的有效处理与可视技术,传统的网络可视化技术对于异质网络可视化来说布局效果混乱、异质信息难以体现。为此提出一种基于动态投影嵌入的多维度异质网络可视化方法。该方法从异质网络的表示学习方法入手,提出动态投影嵌入模型来学习异质网络的节点表示,在此基础上,提出了多维度(空间)的可视化方法,将异质网络节点根据不同属性映射至不同关系空间中进行可视化分析,从而挖掘出潜在的语义信息。实验结果表明,提出的方法不仅使异质网络表示学习的评价指标(MRR)提升了10%,而且从多维度(空间)对异质网络进行可视化,有效地展示和挖掘了网络中的异质信息与潜在语义信息。

关键词: 异质网络, 可视化, 网络表示学习

Abstract:

Heterogeneous network is a complex information network with many types of nodes and edges, so the visuali-
zation of heterogeneous network usually involves effective processing and visualization technology of heterogeneous information. Traditional network visualization methods are confusing in layout effect and difficult to reflect heterogeneous information for heterogeneous network visualization. A multidimensional visualization method of heterogeneous network based on dynamic projection embedding is proposed. This method starts with the representation learning of heterogeneous network, and proposes the dynamic projection embedding algorithm to learn the node representation. On this basis, a multidimensional visualization method is proposed to mine potential semantic information, which maps heterogeneous network node to different relational spaces according to different attributes of node. The experimental results show that the proposedmethod not only improves the MRR of heterogeneous network representation learning by 10%, but also visualizes the heterogeneous network from multiple dimensions, effectively displaying and mining the heterogeneous information and potential semantic information in the network.

Key words: heterogeneous network, visualization, network representation learning