Hierarchical Graph Matching Network for Similarity Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Graph Neural Networks (GNNs) face challenges in effectively computing similarity metrics between arbitrary pairs of graph-structured objects, limiting their application in various fields such as malware detection and natural language processing.
Innovation Solution
A Hierarchical Graph Matching Network (HGMN) is employed, comprising a Siamese Graph Neural Network (SGNN) for global-level interactions and a Multi-perspective Node-Graph Matching Network (NGMN) for cross-level interactions, which learns node embeddings and computes similarity scores through graph-graph classification or regression tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current Graph Neural Networks (GNNs) are used to compute similarity metrics between graph-structured objects, then the computation can be performed, but the accuracy is insufficient and processing efficiency is limited
Solution Approach 1:
The patent segments the graph matching process into multiple hierarchical levels: global-level graph interactions are processed by the Siamese GNN component, while local-level node-graph interactions are processed by the NGMN component. This segmentation allows each component to specialize in specific aspects of graph similarity, improving overall accuracy without requiring a single monolithic model to handle all complexities, thereby maintaining processing efficiency.
Solution Approach 2:
The patent introduces a hierarchical dimension to the graph matching process by considering both global-level and local-level interactions simultaneously. The Siamese GNN operates at the global graph level while the NGMN operates at the local node level, creating a multi-dimensional analysis framework that improves measurement precision without proportionally increasing processing time.
2Measurement precision
If more complex models are used to improve graph similarity computation accuracy, then measurement precision improves, but processing power and storage requirements increase
Solution Approach 1:
The model is segmented into two specialized components: Siamese GNN for global-level graph interaction processing and NGMN for local-level node-graph interaction processing. This segmentation allows each component to be optimized for its specific task, reducing the overall computational burden compared to a single complex model while maintaining high accuracy through coordinated multi-level analysis.
Solution Approach 2:
The hierarchical framework provides multi-functionality by handling both global graph-level similarities and local node-level similarities within a unified architecture. This allows the system to extract diverse features from graph structures efficiently, improving measurement precision without requiring multiple separate models that would increase processing power requirements.
Data Source
AI summary
A computer-implemented method for calculating a similarity between a pair of graph-structured objects by learning-based techniques. The operations include computing the node embeddings of a pair of graph-structured objects of two computer graphs utilizing a hierarchical graph matching network (HGMN). A first component of the HGMN performs graph matching of global-level graph interactions of the two computer graphs. A second component of the HGMN performs graph matching of cross-level node-graph interactions of the two computer graphs. There is an aggregating of features learned from the graph matching of the global-level graph interactions and the cross-level node-graph interactions. At least one of a graph-graph classification or a graph-graph regression is performed utilizing the learned features of the two computer graphs.