一种基于联邦学习的图数据集跨域协同构建方法
By using federated learning and privacy-preserving computation techniques, the attribute completion and structure repair tasks of graph datasets are decoupled, enabling high-quality construction and diversity assessment of cross-institutional graph datasets. This solves the problems of insufficient data value quantification and privacy protection in existing technologies and improves the training effect of graph neural networks.
CN122413482APending Publication Date: 2026-07-17GUIZHOU UNIVERSITY OF FINANCE AND ECONOMICS
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUIZHOU UNIVERSITY OF FINANCE AND ECONOMICS
- Filing Date
- 2026-06-18
- Publication Date
- 2026-07-17
Smart Images
Figure CN122413482A_ABST
Abstract
本发明公开了一种基于联邦学习的图数据集跨域协同构建方法。用于高质量图数据集跨域构建领域，包括：各参与方在本地对图数据集执行多维质量预检，经差分隐私处理后生成质量元数据；基于信息论贡献度评分与属性分布互补性度量对参与方进行双维度撮合准入；采用隐私集合求交协议对节点对齐并对边存在性施加差分隐私保护；将缺失属性语义补全与图结构边修复解耦为两个独立子任务，通过联邦聚合协同完成；最终从节点属性完整性、边关系准确性及拓扑结构合理性进行多维自动化评测并生成报告。本发明在不共享原始图数据的前提下，实现了高质量跨域图数据集的协同构建与标准化评测。
Need to check novelty before this filing date? Find Prior Art