计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (9): 36-49.DOI: 10.3778/j.issn.1002-8331.2012-0410

• 热点与综述 • 上一篇    下一篇

异构图卷积网络研究进展

贾香恩,董一鸿,朱锋,钱江波   

  1. 宁波大学 信息科学与工程学院,浙江 宁波 315211
  • 出版日期:2021-05-01 发布日期:2021-04-29

Research Progress of Heterogeneous Graph Convolutional Networks

JIA Xiang’en, DONG Yihong, ZHU Feng, QIAN Jiangbo   

  1. Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211, China
  • Online:2021-05-01 Published:2021-04-29

摘要:

现实世界中的很多场景都能用多种节点类型和边类型组成的异构网络表示。网络中蕴含着丰富语义关系,并具有实际应用价值,引起了学术界和工业界的关注。传统的方法都是基于浅层模型进行异构网络挖掘。近几年,由于图卷积网络在同构网络中表现优越,有许多学者将图卷积网络应用到异构网络的挖掘,在各个任务中都取得了优异的成绩。通过对异构图卷积网络的研究进展进行评述,来了解相关领域的发展状况。介绍了异构图卷积网络的发展,将异构图卷积网络分为基于元路径和自适应异构信息的模型进行详细介绍及归纳,并综合分析了不同的聚合方法。介绍了异构图卷积网络在推荐系统、生物化学、异常检测和自然语言处理中的应用。分析了异构图卷积网络未来面临的挑战以及值得研究的问题。

关键词: 异构网络, 图卷积网络, 网络表示学习

Abstract:

Many scenarios in the real world can be represented by heterogeneous networks composed of multiple node types and edge types. The networks are rich in semantic relationships and have practical applications that have attracted the attention of academic and industry. Traditional approaches are based on shallow models for heterogeneous network mining. In recent years, due to the superior performance of graph convolutional networks in homogeneous networks, many scholars have applied graph convolutional networks to mining heterogeneous networks and achieved excellent results in various tasks. The research progress of heterogeneous graph convolutional networks is reviewed to understand the development of related fields. The development of heterogeneous graph convolutional networks is introduced, and the heterogeneous graph convolutional networks are divided into models based on meta-path and adaptive heterogeneous information are introduced and summarized in detail, and different aggregation methods are synthesized and analyzed. The applications of heterogeneous graph convolutional networks in recommendation systems, biochemistry, anomaly detection and natural language processing are introduced. The future challenges of heterogeneous graph convolutional networks and the problems worthy of research are analyzed.

Key words: heterogeneous networks, graph convolutional networks, network representation learning