.png){kind=icon}
返回
A lightweight hierarchical graph convolutional model for knowledge graph representation learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-024-05787-2.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Graph convolutional networks (GCNs) have emerged as powerful tools for handling graph-structured data. Many knowledge graph embedding models leverage GCNs as encoders to learn the relationships between central entities and their neighbors, showing impressive performance in the knowledge graph completion task. However, the incorporation of GCNs increases the computational burden of the model. Moreover, due to the complex graph structure of knowledge graphs, treating all neighboring entities equally will cause the model to lose its ability to capture important information. To address these challenges, we present a lightweight hierarchical graph convolutional network (Light-HGCN). Light-HGCN removes feature transformation and selectively uses nonlinearity activation in standard GCNs to accelerate model convergence. Additionally, Light-HGCN introduces a hierarchical attention mechanism to determine the neighboring weights to capture complex graph structures. Light-HGCN achieves promising performance for the link prediction task across multiple benchmark datasets. Ablation experiments further illustrate the effectiveness of the hierarchical attention mechanism. The analysis of feature transformation and nonlinearity activation in GCNs on the KGC task indicates that lightweight GCNs can enhance computational efficiency while preserving promising performance.
Keyword:
Graph convolutional networks simplification
Hierarchical attention mechanism
Knowledge graph representation learning
Knowledge graph completion
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
引用论文
Kinematic analysis of limb movements in neuropsychological research: Subtle deficits and recovery of function.神经心理学研究中肢体运动的运动学分析: 细微的缺陷和功能的恢复。
PREAI
分享收藏
Firing and intrinsic properties of antennal lobe neurons in the Noctuid moth Agrotis ipsilon.png)
Biosystems
OAAI
BiosystemsIF0
OAAI分享收藏DensE: An enhanced non-commutative representation for knowledge graph embedding with adaptive semantic hierarchyNEUROCOMPUTINGOAAI
NEUROCOMPUTINGIF6.5
OAAI分享收藏RAGAT: Relation Aware Graph Attention Network for Knowledge Graph CompletionRAGAT: 用于知识图补全的关系感知图注意力网络IEEE ACCESSOAAI
IEEE ACCESSIF3.6
OAAI分享收藏
查看更多内容