Journal of Computer Applications

    Next Articles

Unsupervised attributed graph embedding model based on node similarity


  • Received:2021-07-14 Revised:2021-09-03 Online:2021-09-15 Published:2021-09-15



  1. (1.东北大学 计算机科学与工程学院,沈阳 110819
    2.北京理工大学 计算机学院,北京 100081
  • 通讯作者: 袁野

Abstract: Attributed graph embedding aims to transform the nodes in an attributed graph into low dimension vectors while preserving the topology and attributes of the node. There are lots of related work in attributed graph embedding. Most of them are supervised or semi-supervised. In practical applications, the number of nodes that need to be labeled is large, which is difficult and consumes huge manpower and material resources. Above problems from an unsupervised perspective were reanalyzed and an unsupervised attributed graph embedding algorithm was proposed. The topology information and attribute information of the nodes respectively was calculated by using the existing pure graph embedding algorithm and attributes firstly. Then the embedding of the nodes was obtained by using Graph Convolutional Network (GCN), and the difference between embedding and the topology information and the difference between embedding and attribute information was minimized to ensure that finally paired nodes with similar topological information and attribute information got similar embeddings. Compared with Graph Auto-Encoder (GAE), the node classification accuracy of the proposed method on three data sets is improved by 1.2%, 2.4% and 0.9% respectively. Experimental results show that the proposed method can effectively improve the quality of the generated embedding.

Key words: attributed graph embedding, Graph Convolution Network (GCN), node classification, node similarity, unsupervised

摘要: 属性图嵌入旨在将属性图中的节点表示为低维向量,同时保留节点的拓扑信息和属性信息。属性图嵌入已经有一系列相关工作,它们大多数是有监督或半监督。在实际应用中,需要标记的节点数量多,难度大,消耗了巨大的人力物力。本文针对上述问题以无监督的视角重新分析,提出了一种无监督的属性图嵌入算法。该算法首先通过已存在的无属性图嵌入算法和图的属性分别计算节点的拓扑信息和属性信息,其次利用图卷积网络(GCN)得到节点的嵌入向量,并使得嵌入向量与拓扑信息及属性信息的差最小,最终使拓扑信息和属性信息都相似的成对节点得到相似嵌入。与图自动编码器(GAE)相比,本文所提出的方法在三个数据集上的节点分类准确率分别提升了1.2%,2.4%,0.9%。实验结果表明,本文所提出方法能够有效提高生成的嵌入的质量。

关键词: 属性图嵌入, 图卷积网络, 节点分类, 节点相似度, 无监督

CLC Number: