Title:
METHOD AND DEVICE FOR ACQUIRING DYNAMIC EMBEDDING VECTOR OF NODE IN RELATIONAL NETWORK DIAGRAM
Document Type and Number:
WIPO Patent Application WO/2020/224298
Kind Code:
A1
Abstract:
A method and device for obtaining a dynamic embedding vector of a node in a relational network diagram of a plurality of nodes. The method comprises: determining N neighbor nodes of a first node on the basis of a relational network diagram of a plurality of nodes at a first moment (S202); obtaining input embedding vectors of the first node and the N neighbor nodes thereof (S204); inputting the input embedding vectors of the first node and the N neighbor nodes thereof into a pre-trained embedding model, to enable the model to output: in a case where the first node is a main node, dynamic embedding vectors of the first node and the plurality of neighbor nodes thereof (S206), wherein the embedding model comprises at least one calculation module which is connected front and back, each calculation module comprises a corresponding self-attention layer, and the self-attention layer outputs N + 1 layer output vectors corresponding to N + 1 input vectors respectively on the basis of a self-attention mechanism with respect to the N + 1 input vectors of the self-attention layer.
Inventors:
CAO SHAOSHENG (CN)
CUI QING (CN)
CUI QING (CN)
Application Number:
PCT/CN2020/071704
Publication Date:
November 12, 2020
Filing Date:
January 13, 2020
Export Citation:
Assignee:
ADVANCED NEW TECHNOLOGIES CO LTD (CN)
International Classes:
G06F16/901
Foreign References:
| CN110245269A | 2019-09-17 | |||
| CN108985309A | 2018-12-11 | |||
| CN107145977A | 2017-09-08 | |||
| CN108737491A | 2018-11-02 | |||
| CN108256678A | 2018-07-06 | |||
| US20170337481A1 | 2017-11-23 |
Attorney, Agent or Firm:
BEIJING BESTIPR INTELLECTUAL PROPERTY LAW CORPORATION (CN)
Download PDF: