Latent Network Summarization for Graph Storage Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional graph representation techniques, such as node embedding and graph summarization, face challenges with large-scale graphs due to high storage requirements and computational inefficiencies, particularly in handling billions of nodes, and often sacrifice information fidelity for space efficiency, limiting their applicability to specific tasks and preventing on-the-fly node embedding derivation.
Innovation Solution
The latent summarization method learns a compressed, multi-level structural representation of graphs using relational operators and functions, capturing structural information at various levels, which is independent of graph size, allowing for efficient storage and on-the-fly node embedding derivation, supporting diverse graph-based tasks without tailored approaches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of stationary object
If conventional graph representation techniques (node embedding, graph summarization) are used to represent large-scale graphs, then storage space is reduced, but information fidelity is lost and computational efficiency deteriorates
Solution Approach 1:
The patent segments the graph representation into multiple levels of feature matrices (base feature matrix, first-level feature matrix, second-level feature matrix, etc.), where each level captures structural information at different granularities. This segmentation allows the system to store only essential structural patterns rather than complete graph data, reducing storage while preserving information fidelity through hierarchical abstraction.
Solution Approach 2:
The patent transforms the graph from traditional node-embedding representation into a multi-level feature matrix representation that adds a dimensional hierarchy. Instead of flattening the graph into fixed-dimensional vectors, the system creates matrices organized by structural levels, enabling efficient storage through dimensionality reduction while maintaining the ability to derive node embeddings on-the-fly when needed.
2Volume of stationary object
If conventional graph representation techniques are used for large-scale graphs, then storage requirements are reduced, but computational efficiency deteriorates
Solution Approach 1:
The patent performs preliminary computation to extract and store multi-level structural features in feature matrices before actual graph tasks are executed. By pre-computing and storing these hierarchical structural representations, the system enables fast on-the-fly derivation of node embeddings during runtime without reprocessing the entire graph, thus improving computational efficiency while maintaining reduced storage requirements.
Solution Approach 2:
The patent creates compressed copies of graph structural information in the form of feature matrices that can be stored efficiently. These matrices serve as compact representations that can be quickly loaded and processed, replacing the need to store and process the complete original graph structure, thereby improving computational efficiency while maintaining storage efficiency.
3Volume of stationary object
If conventional graph summarization is applied to achieve space efficiency, then storage is reduced, but adaptability to diverse graph tasks is limited
Solution Approach 1:
The patent creates a universal multi-level feature matrix representation that can serve multiple graph-based tasks. The hierarchical structure captures structural information at different levels of abstraction, making the representation adaptable to various tasks such as node classification, link prediction, and community detection without requiring task-specific tailored approaches, thus achieving both storage efficiency and task versatility.
Solution Approach 2:
The patent enables parameter changes by allowing flexible selection of which feature matrices to use and how to combine them based on task requirements. The system can adjust the level of detail and structural information captured by modifying which hierarchical levels are utilized, providing adaptability to different graph tasks while maintaining the compact stored representation.
4Reliability
If conventional node embedding is used, then graph structure is represented, but storage space increases and on-the-fly derivation is prevented
Solution Approach 1:
The patent extracts essential structural information from the complete graph and stores it in compressed multi-level feature matrices. Instead of storing full node embeddings for all nodes, the system extracts and stores only the critical structural patterns in hierarchical form, enabling space-efficient storage while maintaining the ability to derive complete node embeddings on-the-fly when needed for specific tasks.
Data Source
AI summary
Embodiments of the present invention provide systems, methods, and computer storage media for latent summarization of a graph. Structural features can be captured from feature vectors associated with each node of the graph by applying base functions on the feature vectors and iteratively applying relational operators to successive feature matrices to derive deeper inductive relational functions that capture higher-order structural information in different subgraphs of increasing size (node separations). Heterogeneity can be summarized by performing capturing features in appropriate subgraphs (e.g., node-centric neighborhoods associated with each node type, edge direction, and/or edge type). Binning and/or dimensionality reduction can be applied to the resulting feature matrices. The resulting set of relational functions and multi-level feature matrices can form a latent summary that can be used to perform a variety of graph-based tasks, including node classification, node clustering, link prediction, entity resolution, anomaly and event detection, and inductive learning tasks.