Title:
GRAPH NEURAL NETWORK TRAINING METHOD AND SYSTEM, AND ABNORMAL ACCOUNT IDENTIFICATION METHOD
Document Type and Number:
WIPO Patent Application WO/2024/087844
Kind Code:
A1
Abstract:
Provided in the present disclosure are a graph neural network training method and system, and an abnormal account identification method. The graph neural network training method comprises: acquiring initial graph structure data corresponding to a terminal device, wherein initial graph structure data acquired by each of a plurality of distributed training terminals comes from the same sample graph structure data; and cyclically executing the following graph structure data processing stage and graph neural network training stage until a target neural network meeting training requirements is obtained: determining a processing occasion of the current instance of execution of the graph structure data processing stage according to historical execution data of historical instances of execution of the graph structure data processing stage and the graph neural network training stage; according to the processing occasion, performing graph structure data processing on the initial graph structure data in the graph structure data processing stage, so as to generate target graph structure data, wherein the graph structure data processing comprises data sampling processing and feature extraction processing; and in the graph neural network training stage, training the target neural network on the basis of the target graph structure data.
More Like This:
Inventors:
CHEN HONGZHI (CN)
Application Number:
PCT/CN2023/114920
Publication Date:
May 02, 2024
Filing Date:
August 25, 2023
Export Citation:
Assignee:
BEIJING VOLCANO ENGINE TECH CO LTD (CN)
International Classes:
G06N3/08
Domestic Patent References:
| WO2022141869A1 | 2022-07-07 |
Foreign References:
| CN114936637A | 2022-08-23 | |||
| CN114330670A | 2022-04-12 |
Attorney, Agent or Firm:
SHIHUI PARTNERS (CN)
Download PDF: