添加链接
link管理
链接快照平台
  • 输入网页链接,自动生成快照
  • 标签化管理网页链接
Gong Pei-Liang, Ai Li-Hua. Two-order approximate spectral convolutional model for semi-supervised classification. Acta Automatica Sinica, 2021, 47(5): 1067−1076 doi: 10.16383/j.aas.c200040 Citation: Gong Pei-Liang, Ai Li-Hua. Two-order approximate spectral convolutional model for semi-supervised classification. Acta Automatica Sinica, 2021, 47 (5): 1067−1076 doi: 10.16383/j.aas.c200040 Gong Pei-Liang, Ai Li-Hua. Two-order approximate spectral convolutional model for semi-supervised classification. Acta Automatica Sinica, 2021, 47(5): 1067−1076 doi: 10.16383/j.aas.c200040 Citation: Gong Pei-Liang, Ai Li-Hua. Two-order approximate spectral convolutional model for semi-supervised classification. Acta Automatica Sinica, 2021, 47 (5): 1067−1076 doi: 10.16383/j.aas.c200040 作者简介:

公沛良:北京交通大学计算机与信息技术学院硕士研究生. 主要研究方向为图数据分析, 数据挖掘, 机器学习和认知计算. E-mail: [email protected]

艾丽华:博士, 北京交通大学计算机与信息技术学院副教授. 主要研究方向为大型图数据挖掘, 神经网络计算, 机器学习, 并行计算和分布式计算. 本文通信作者.E-mail: [email protected]

Funds: Supported by National Natural Science Foundation of China (61472029, 51827813, 61473031)
More Information Author Bio: GONG Pei-Liang Master student at the School of Computer and Information Technology, Beijing Jiaotong University. His research interest covers graph data analysis, data mining, machine learning, and cognitive computing

AI Li-Hua Ph.D., associate professor at the School of Computer and Information Technology, Beijing Jiaotong University. Her research interest covers large-scale graph mining, neural network computing, machine learning, parallel computing, and distributed computing. Corresponding author of this paper

近年来, 基于局部一阶近似的谱图卷积方法在半监督节点分类任务上取得了明显优势, 但是在每次更新节点特征表示时, 只利用了一阶邻居节点信息而忽视了非直接邻居节点信息. 为此, 本文结合切比雪夫截断展开式及标准化的拉普拉斯矩阵, 通过推导及简化二阶近似谱图卷积模块, 提出了一种融合丰富局部结构信息的改进图卷积模型, 进一步提高了节点分类性能. 大量的实验结果表明, 本文提出的方法在不同数据集上的表现均优于现有的流行方法, 验证了模型的有效性.

图理论 /  谱图卷积 /  半监督学习 /  节点分类 / Abstract:

In recent years, the spectral convolution method based on local first-order approximation has achieved significant advantages in semi-supervised node classification tasks. However, when updating the node feature representation at each stage, only the first-order neighbor node information is used, while the indirect neighbor node information is ignored. To this end, this paper combines Chebyshev′ s truncated expansion and symmetric normalized Laplacian matrix, and by deducing and simplifying the two-order approximate spectral convolution module, an improved graph convolution model is proposed which fuses rich local structure information. A large number of experimental results show that the method proposed in this paper is superior to the existing popular methods on different datasets, which verifies the effectiveness of the model.

Key words: Graph theory /  spectral convolution /  semi-supervised learning /  node classification /  relational data

地址:北京中关村东路95号 邮政编码:100190 E-mail: [email protected]

电话:010-82544677 (日常咨询和稿件处理), 010-82544653(费用管理、寄刊)

北京仁和汇智信息技术有限公司 开发 技术支持: [email protected]