电子学报 ›› 2020, Vol. 48 ›› Issue (10): 2047-2059.DOI: 10.3969/j.issn.0372-2112.2020.10.024

• 综述评论 • 上一篇    下一篇

动态网络表示学习研究进展

曹燕, 董一鸿, 邬少清, 陈华辉, 钱江波, 潘善亮   

  1. 宁波大学信息科学与工程学院, 浙江宁波 315211
  • 收稿日期:2019-07-01 修回日期:2020-03-02 出版日期:2020-10-25
    • 通讯作者:
    • 董一鸿
    • 作者简介:
    • 曹燕 女,1993年出生.CCF学生会员,宁波大学信息科学与工程学院硕士,主要研究方向为大数据、数据挖掘.
    • 基金资助:
    • 浙江省自然科学基金 (No.LY20F020009,No.LZ20F020001); 国家自然科学基金 (No.61502133)

Dynamic Network Representation Learning:A Review

CAO Yan, DONG Yi-hong, WU Shao-qing, CHEN Hua-hui, QIAN Jiang-bo, PAN Shan-liang   

  1. Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211, China
  • Received:2019-07-01 Revised:2020-03-02 Online:2020-10-25 Published:2020-10-25
    • Corresponding author:
    • DONG Yi-hong
    • Supported by:
    • National Natural Science Foundation of Zhejiang Province,  China (No.LY20F020009, No.LZ20F020001); National Natural Science Foundation of China (No.61502133)

摘要: 网络表示学习旨在将网络信息表示为低维稠密的实数向量,解决链接预测、异常检测、推荐系统等任务.近年来,网络表示学习研究取得重大进展,但研究多基于静态网络,而真实世界构成的网络是动态变化的,对动态网络分析的需求日益增加.本文总结了当前动态网络表示学习的方法与研究进展,首先提出网络表示学习的动机,阐述动态网络以及表示学习的发展历史与理论基础;接着,系统概述了大量动态网络嵌入方法,包括基于矩阵分解的动态图嵌入、基于随机游走的动态图嵌入、基于深度学习的动态图嵌入和基于重构概率的动态图嵌入,并分析与比较,给出动态网络表示学习的应用场景;最后,总结未来网络表示学习的研究方向.只有考虑网络的动态性,才能真实反映现实网络的演化,使网络表示学习更具价值.

关键词: 网络表示学习, 图嵌入, 深度学习, 图卷积网络

Abstract: Network representation learning aims to learn the low-dimensional dense real-valued vector of network information, which solves practical tasks such as link prediction, anomaly detection, and recommendation systems. Recently, network representation learning has made great progress. Most existing researches focus on static networks, while real network is dynamic all the time. This survey proposes state of the arts on representation learning of dynamic network. Firstly, it provides historical overview of representation learning in network, followed by the motivation and theoretical basis of dynamic network representation learning. Then comprehensive analysis of dynamic models is proposed, including matrix factorization, random-walk, deep learning, edge reconstruction based dynamic models, and gives the application scenarios of dynamic network embedding. Finally, research directions of representation learning in the future are summarized. Only when considering the temporal dynamics, structure and content can we truly reflect the evolution of the real network and make network representation learning more valuable.

Key words: network representation learning, graph embedding, deep learning, graph convolutional networks

中图分类号: