基于二部联合网络的属性缺失图学习方法

打开文本图片集

中图分类号：TP18;O157.5文献标识码：A

Abstract：Aiming at the problem of missing node attributes in graph data，we proposes a novel attribute missing graph learning framework. The framework maps node attributes to edge attributes by reconstructing the structural joint bipartite network. This enables attribute completion and graph tasks to be performed together under a unified framework that can handle both continuous and discrete missing data. According to the attribute homogeneity and structural homogeneity of the attribute graph，we propose an attribute missing representation learning method，which introduces edge embeddings and attention mechanisms to capture the correlations between attribute-attribute and structure-attribute in structural joint bipartite network to enhance the missing attribute learning. Experimental results on four real-world datasets show that our framework outperforms the baselines in both atribute completion tasks，validating the effectiveness of the method.

Keywords： graph neural network；attribute completion; node classification; bipartite graph;topology of networks

0 引言

图结构数据可以自然地模拟来自社会[1-2]、金融[3-4]、生物[5]等领域的真实数据。（剩余12959字）