.png){kind=icon}
.png){kind=icon}
Return
MultiRelE: Multi-relation Knowledge Graph Embedding Model
.png){kind=icon}
PRE
AI
DOI：10.1007/978-981-95-5719-6_22.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In order to solve the problems that knowledge graph embedding models increase the relationship size proportionally with the increase of entity dimension, resulting in a surge in the number of parameters, and it is difficult to identify relationship patterns in higher dimensions, this paper proposed a multi-relational knowledge graph embedding model, called MultiRelE. MultiRelE model uses matrix to represent entities and Kronecker product orthogonal matrix to represent relationships. It combines singular value thresholding and Grassmann manifold optimization to shrink the relationship matrix to reduce the number of parameters. Experimental results show that on WN18RR and FB15K-237 datasets, MultiRelE captures several relational patterns, and shows significant advantages over the suboptimal model DCNE while significantly reducing the number of parameters, with MRR increased by 7.72%and 8.19%respectively. The results show that the MultiRelE model has better indicators in dealing with multi-relational knowledge graphs and high-dimensional data scenarios, and provides a new solution for the field of knowledge graph embedding.
Keywords:
Knowledge graph embedding
High-dimensional rotation
Multi-relation
Matrix representation
Contracting Grassmann Manifold Optimization
Journal
W
IF:
0
Papers:
32
Citations:
0