Distributed graph embedding-based federated graph clustering method, apparatus, and readable storage medium
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current federated learning technologies lack effective privacy-compliant graph mining algorithms for multi-party data, particularly in graph computing, which involves complex topology interactions and lacks efficient methods for mining organized behaviors.
Innovation Solution
A distributed graph embedding-based federated graph clustering method that constructs graphs from party data, performs encrypted intersection to align common nodes, and uses a random walk-based algorithm to learn graph embedding vectors, followed by clustering analysis to obtain results while maintaining data privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If graph computing is performed on multi-party data to mine organized behaviors, then the value of multi-party data can be mined, but the computational complexity increases significantly due to multiple rounds of topology interaction computation
Solution Approach 1:
The patent segments the graph clustering task into two independent phases: (1) local graph embedding learning on each party's data separately, and (2) federated clustering on the embedded representations. This segmentation avoids the need for complex multi-party topology interactions while preserving the ability to mine organized behaviors from the combined data structure.
Solution Approach 2:
The patent introduces graph embedding vectors as an intermediary representation that captures the topological structure and node attributes. Instead of performing direct graph computations on the original multi-party data, the system operates on these embedded vectors, which serve as a simplified mediator that preserves essential structural information while reducing computational complexity.
2Reliability
If conventional federated learning algorithms are used for individual value portrait evaluation, then data privacy is preserved, but the mining of potential organized behaviors is comparatively lacking
Solution Approach 1:
The patent changes the parameter representation from raw multi-party data to graph embedding vectors that capture topological relationships. This parameter transformation enables the model to detect organized behaviors and structural patterns that conventional federated learning algorithms miss, while still operating in a privacy-preserving federated framework.
Solution Approach 2:
The patent transitions from analyzing data in the original feature space to analyzing graph embedding vectors in a transformed dimensional space that explicitly encodes topological relationships. This dimensional change allows the system to mine organized behaviors by capturing node positions, connectivity patterns, and community structures that are not apparent in the original data representation.
3Loss of information
If graph structures are processed in their original form, then complete topological information is preserved, but the computational efficiency of federated graph calculations decreases
Solution Approach 1:
The patent extracts the essential topological information from the complete graph structure and encodes it into compact graph embedding vectors. This extraction process separates the critical structural features from the full graph data, enabling efficient federated computation while preserving the topological relationships needed for meaningful clustering and behavior mining.
Data Source
AI summary
Provided are a federated graph clustering method based on distributed graph embedding, a device, and a readable storage medium. The method comprises: constructing a first graph on the basis of first party data, and constructing a second graph on the basis of second party data; performing an encrypted intersection on the first party data and the second party data, determining common nodes in the first graph and the second graph, and associating the first graph with the second graph according to the common nodes to obtain a federated graph; learning on the federated graph by using a random walk-based distributed graph embedding algorithm, and determining a first graph embedding vector [PiA, PiB] starting from the first graph and a second graph embedding vector [PiA′, PiB′] starting from the second graph; and performing a clustering analysis on the first graph embedding vector [PiA, PiB] and the second graph embedding vector [PiA′, PiB′] of the federated graph on the basis of a federated clustering method to obtain a clustering result. By utilizing the present method, federated graph clustering can be carried out on private data of two parties, and a better clustering effect is obtained.