METHOD FOR FUSING KNOWLEDGE GRAPH BASED ON ITERATIVE COMPLETION

WIPO Patent Application WO/2022/011681

A1

A method for fusing a knowledge graph based on iterative completion, comprising the following steps: acquiring multiple items of knowledge graph data, and identifying all of the entities in the knowledge graph; acquiring a structural vector and an entity name vector of each entity; calculating the structural similarity and entity name similarity between entities; establishing a mutual attention network based on degree perception, and calculating the entity similarity between entities after fusion; on the basis of the entity similarity, selecting an entity pair with a high degree of confidence, and using iterative training to perform knowledge graph completion to acquire a fused knowledge graph. The method of the present invention proposes a degree-perception mutual attention network to fuse entity name vectors and structural vectors in order to make the alignment effect better, proposes using a spliced power average word vector to characterise entity names, which can capture more information of the entity names, and proposes an iterative training algorithm based on knowledge graph completion to iteratively enhance the entity alignment effect so that long-tail entities are more easily aligned.

ZHAO XIANG (CN)
ZENG WEIXIN (CN)
TANG JIUYANG (CN)
LI XINYI (CN)
TAN ZHEN (CN)
GUO DEKE (CN)
LUO LAILONG (CN)

PCT/CN2020/102683

January 20, 2022

July 17, 2020

Click for automatic bibliography generation  

NATIONAL UNIV OF DEFENSE TECHNOLOGY (CN)

G06F16/28; G06F16/36

CN111291139A	2020-06-16
CN110489554A	2019-11-22
CN110851613A	2020-02-28
CN110245299A	2019-09-17
CN111339313A	2020-06-26
US20200104311A1	2020-04-02

LECOME INTELLECTUAL PROPERTY AGENT LTD. (CN)

View/Download PDF      PDF Help

Previous Patent: ONE-PIECE CIGARETTE

Next Patent: WIRELESS COMMUNICATION METHOD, TERMINAL DEVICE, AND NETWORK DEVICE