《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (1): 9-15.DOI: 10.11772/j.issn.1001-9081.2021071289

• 人工智能 • 上一篇    下一篇

利用初始残差和解耦操作的自适应深层图卷积

张继杰1, 杨艳1,2(), 刘勇1,2   

  1. 1.黑龙江大学 计算机科学技术学院,哈尔滨 150080
    2.黑龙江省数据库与并行计算重点实验室(黑龙江大学),哈尔滨 150080
  • 收稿日期:2021-07-19 修回日期:2021-08-13 接受日期:2021-08-19 发布日期:2021-08-13 出版日期:2022-01-10
  • 通讯作者: 杨艳
  • 作者简介:张继杰(1998—),男,山东青岛人,硕士研究生,CCF会员,主要研究方向:图表示学习、推荐系统
    杨艳(1973—),女,黑龙江哈尔滨人,教授,博士,CCF会员,主要研究方向:数据挖掘、社会网
    刘勇(1975—),男,河北昌黎人,副教授,博士,CCF会员,主要研究方向:数据挖掘、机器学习。

Adaptive deep graph convolution using initial residual and decoupling operations

Jijie ZHANG1, Yan YANG1,2(), Yong LIU1,2   

  1. 1.School of Computer Science and Technology,Heilongjiang University,Harbin Heilongjiang 150080,China
    2.Key Laboratory of Database and Parallel Computing of Heilongjiang Province (Heilongjiang University),Harbin Heilongjiang 150080,China
  • Received:2021-07-19 Revised:2021-08-13 Accepted:2021-08-19 Online:2021-08-13 Published:2022-01-10
  • Contact: Yan YANG
  • About author:ZHANG Jijie, born in 1998, M. S. candidate. His research interests include graph representation learning, recommendation system.
    YANG Yan, born in 1973, Ph. D., professor. Her research interests include data mining, social network.
    LIU Yong, born in 1975, Ph. D., associate professor. His research interests include data mining, machine learning.
  • Supported by:
    Natural Science Foundation of Heilongjiang Province(LH2020F043)

摘要:

传统的图卷积网络(GCN)及其很多变体都是在浅层时达到最佳的效果,而没有充分利用图中节点的高阶邻居信息。随后产生的深层图卷积模型可以解决以上问题却又不可避免地产生了过平滑的问题,导致模型无法有效区分图中不同类别的节点。针对此问题,提出了一种利用初始残差和解耦操作的自适应深层图卷积模型ID-AGCN。首先,对节点的表示转换以及特征传播进行解耦;然后,在节点的特征传播过程中添加了初始残差;最后,自适应地结合不同传播层得到的节点表示,针对每个节点选择其合适的局部信息和全局信息以得到含有丰富信息的节点表征,并利用少部分带标签的节点进行监督训练来生成最终的节点表征。在Cora、CiteSeer和PubMed这三个数据集上的实验结果表明,ID-AGCN的分类准确率相较GCN分别提高了约3.4个百分点、2.3个百分点和1.9个百分点。所提模型能够更好地缓解过平滑。

关键词: 节点分类, 初始残差, 解耦, 自适应, 图卷积网络

Abstract:

The traditional Graph Convolutional Network (GCN) and many of its variants achieve the best effect in the shallow layers, and do not make full use of higher-order neighbor information of nodes in the graph. The subsequent deep graph convolution models can solve the above problem, but inevitably generate the problem of over-smoothing, which makes the models impossible to effectively distinguish different types of nodes in the graph. To address this problem, an adaptive deep graph convolution model using initial residual and decoupling operations, named ID-AGCN (model using Initial residual and Decoupled Adaptive Graph Convolutional Network), was proposed. Firstly, the node’s representation transformation as well as feature propagation was decoupled. Then, the initial residual was added to the node’s feature propagation process. Finally, the node representations obtained from different propagation layers were combined adaptively, appropriate local and global information was selected for each node to obtain node representations containing rich information, and a small number of labeled nodes were used for supervised training to generate the final node representations. Experimental result on three datasets Cora, CiteSeer and PubMed indicate that the classification accuracy of ID-AGCN is improved by about 3.4 percentage points, 2.3 percentage points and 1.9 percentage points respectively, compared with GCN. The proposed model has superiority in alleviating over-smoothing.

Key words: node classification, initial residual, decoupling, adaptive, Graph Convolutional Network (GCN)

中图分类号: