Multilevel graph comparison enhancement method and system based on homogeneity hypothesis, and terminal

CN120259694APending Publication Date: 2025-07-04SHENZHEN UNIV
0 Cites -1 Cited by

Patent Information

Application Number
CN202510175615.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-02-18
Publication Date
2025-07-04

AI Technical Summary

Technical Problem

In the prior art, the comparative learning method relies on the original graph to construct data pairs, ignoring the defects in the graph structure, resulting in weak homogeneity of the graph and low classification accuracy of data nodes.

Method used

Through a multi-level graph comparison enhancement method based on the homogeneity assumption, the feature matrix and adjacency matrix of the initial graph are constructed, graph diffusion processing is performed, the target diffusion matrix and adjacency set are generated, the target adjacency matrix is ​​constructed, the positive sample set of the multi-level comparison model is obtained, and the optimization loss function is represented by orthogonal constraints and mutual information is used to represent the optimization loss function, and the multi-level comparison model is trained.

Benefits of technology

It improves the accuracy of node classification, enhances the structural information and feature embedding of the graph, reduces noise edges, and improves the accuracy of the classification results of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120259694A_ABST
    Figure CN120259694A_ABST
Patent Text Reader

Abstract

The invention discloses a homogeneity hypothesis-based multi-level graph comparison enhancement method, system and terminal, and the method comprises the steps: constructing a feature matrix and an adjacent matrix of a graph according to a data set, constructing a high-quality graph based on a structure and features, and then collecting positive samples at node, local and global levels, so as to carry out multi-level comparison learning; and two regular terms are introduced to train the model, so that the distinguishability represented by each node in a data set is enhanced, and the node classification accuracy is improved. According to the method, homogeneity hypothesis is utilized to generate more homogeneous edges so as to refine an input graph and further enhance structural information, two regularization items are introduced to enrich feature embedding and realize feature decorrelation, and the accuracy of a classification result output by a model is improved.
Need to check novelty before this filing date? Find Prior Art