Title:
NODE CLASSIFICATION METHOD AND SYSTEM BASED ON TENSOR GRAPH CONVOLUTIONAL NETWORK
Document Type and Number:
WIPO Patent Application WO/2023/221275
Kind Code:
A1
Abstract:
A node classification method and system based on a tensor graph convolutional network. The method comprises: generating a plurality of single-angle-of-view enhanced graphs of different angles of view according to graph data, and performing a tensorization operation according to the graph data and the plurality of obtained single-angle-of-view enhanced graphs, so as to obtain a multi-angle-of-view enhanced graph, wherein the graph data comprises adjacency matrices and feature matrices of graphs, each feature matrix being composed of feature vectors of all nodes in a graph; and the multi-angle-of-view enhanced graph comprises an adjacency tensor and a feature tensor, each front tangent plane of the adjacency tensor representing the adjacency matrix of the single-angle-of-view enhanced graph of a corresponding angle of view, and each front tangent plane of the feature tensor representing the feature matrix of the single-angle-of-view enhanced graph of a corresponding angle of view; and inputting the adjacency tensor and feature tensor of the multi-angle-of-view enhanced graph into a trained tensor graph convolutional network model, and outputting a classification prediction result of each node in the graph data. In the present invention, the robustness of a tensor graph convolutional network model is improved in terms of graph data preprocessing and the model architecture, such that the tensor graph convolutional network model is still relatively stable when suffering from an adversarial attack.
Inventors:
CHEN CHUAN (CN)
WU ZHEBIN (CN)
ZHENG ZIBIN (CN)
WU ZHEBIN (CN)
ZHENG ZIBIN (CN)
Application Number:
PCT/CN2022/105119
Publication Date:
November 23, 2023
Filing Date:
July 12, 2022
Export Citation:
Assignee:
UNIV SUN YAT SEN (CN)
International Classes:
G06K9/62
Foreign References:
| CN108280398A | 2018-07-13 | |||
| CN112862092A | 2021-05-28 | |||
| CN109242007A | 2019-01-18 | |||
| CN106934359A | 2017-07-07 | |||
| CN112488185A | 2021-03-12 | |||
| US20110261050A1 | 2011-10-27 |
Attorney, Agent or Firm:
UNITALEN ATTORNEYS AT LAW (CN)
Download PDF: