Tensor Graph Neural Network for Time-Varying Graph Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing time-varying graphs using tensor graph neural networks lack efficiency and fail to effectively capture correlations over time, particularly in dynamic graph structures.
Innovation Solution
The method involves constructing adjacency and feature tensors from time-varying graphs, partitioning them into training and validation tensors, and creating embedding and prediction models to generate optimized model pairs for predicting future network behaviors, such as social network relationships or business transaction trustworthiness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If tensor graph neural networks are used to analyze time-varying graphs, then the analysis can capture correlations over time, but the existing methods lack efficiency and fail to effectively capture correlations in dynamic graph structures
Solution Approach 1:
The patent segments the time-varying graph analysis into discrete time steps, creating a sequence of static graphs at different time points. This allows the application of efficient static graph neural network algorithms to each time step while capturing temporal correlations through the sequence structure, thereby improving both accuracy and efficiency.
Solution Approach 2:
The patent dynamically updates graph representations at each time step, allowing the model to adapt to changing graph structures while maintaining computational efficiency. The dynamic graph neural network processes graphs at different time points sequentially, capturing temporal patterns without requiring complex simultaneous analysis of all time points.
2Reliability
If dynamic graph neural networks are used to analyze time-varying graphs, then temporal correlations can be captured, but computational efficiency deteriorates
Solution Approach 1:
The patent divides the dynamic graph analysis into separate time steps, processing each time point independently with standard graph neural network algorithms. This segmentation approach captures temporal correlations through the sequence of processed graphs while avoiding the computational burden of analyzing all time points simultaneously in a single complex operation.
Solution Approach 2:
The patent performs preliminary processing at each time step by constructing graph representations and computing embeddings independently before performing temporal correlation analysis. This preliminary action at each time point allows for efficient processing while capturing temporal patterns through subsequent aggregation and prediction operations.
Data Source
AI summary
A computer-implemented method for analyzing a time-varying graph is provided. The time-varying graph includes nodes representing elements in a network, edges representing transactions between elements, and data associated with the nodes and the edges. The computer-implemented method includes constructing, using a processor, adjacency and feature matrices describing each node and edge of each time-varying graph for stacking into an adjacency tensor and describing the data of each time-varying graph for stacking into a feature tensor, respectively. The adjacency and feature tensors are partitioned into adjacency and feature training tensors and into adjacency and feature validation tensors, respectively. An embedding model and a prediction model are created using the adjacency and feature training tensors. The embedding and prediction models are validated using the adjacency and feature validation tensors to identify an optimized embedding-prediction model pair.