.png){kind=icon}
.png){kind=icon}
Return
Graph-aware tensor factorization convolutional network for knowledge graph completion
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-023-01995-3.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Constructed by millions of triples, knowledge graph is a commonly used structured representation of information encoding both the entities of the real-world facts and their corresponding relations. Although knowledge graph contains rich information, it is often far from complete. Knowledge graph completion algorithms are able to fill in the missing information efficiently by inferring missing or correcting wrong facts based on existing ones. However, as a kind of graph structure data, the knowledge graph includes not only the attributes of entities and relationships but also the entire graph structure information. Graph convolutional networks can learn better embedding of entities and relationships by using the connectivity structure of graphs, which is conducive to the knowledge graph complement. Although the existing methods based on graph convolutional networks can simultaneously capture the graph structure information and the attributes of entities and relationships from the knowledge graph, there is still a lack of better information decoupling and representation methods. In this article, we propose a novel Graph-Aware Tensor Factorization Convolutional Network (GATFCN) for knowledge graph completion. GATFCN combines a graph convolutional network and a tucker decomposition as the encoder and decoder. It leverages the advantages of graph convolutional network and tensor decomposition in an end-to-end manner to enrich the representation of facts. For the encoder, the GCN model is used to utilize the local information of the graph structure and graph nodes. For the decoder, tensor factorization can decode the information encoded by GCN while enriching the data representation. According to the comprehensive numerical experiments performed, GATFCN outperforms previous state-of-the-art methods across a number of different benchmark datasets, illustrating its effectiveness in knowledge graph completion.
Keywords:
Knowledge graph completion
Graph convolutional network
Tensor factorization
Journal
IF:
2.7
Papers:
3.1K
Citations:
5.6K