.png){kind=icon}
.png){kind=icon}
Return
Knowledge based attribute completion for heterogeneous graph node classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2024.129023.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Heterogeneous graphs, with diverse node and edge types, are prevalent in real-world scenarios. Graph Neural Networks have gained significant attention for processing such complex data structures, delivering impressive results across various tasks like node classification. However, current GNN approaches overlook quality concerns within heterogeneous graphs, like noise or missing attributes, which directly affect the performance of downstream tasks. Previous research has attempted to address these challenges by employing simple one- hot embedding, which is disconnected from the original graph and prone to introducing noise. However, the improvements brought by these methods are limited when the quality of the original graph itself is poor. To this end, we propose a novel attribute completion method based on external knowledge bases, which incorporates a knowledge based completion framework with a relation-aware attention mechanism. The model first employs an attention mechanism to embed the external knowledge subgraph related to the topological structure of the original graph. Then, for nodes requiring attribute completion, the model leverages the association between the external knowledge base and the original graph to complete the missing attributes. Our proposed method can be combined with existing heterogeneous graph neural networks to enhance their performance. Extensive node classification experiments on three real-world datasets underscore the superior performance of our proposed method.
Keywords:
External knowledge bases
Attribute completion
Node classification
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W