Graph Decomposition for Fraud Subgraph Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional graph building techniques fail to efficiently identify relevant subgraphs within large and connected graphs, making it challenging for fraud analysts to navigate and analyze data effectively, especially in highly connected networks like retail banking systems.
Innovation Solution
The proposed solution involves a graph decomposition technique that uses user-defined interest functions to extract relevant subgraphs by propagating interest across the network, employing a message-passing algorithm to compute expansions, and performing parallel expansion to scale to large networks, thereby providing manageable and context-rich subgraphs for fraud analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional graph building techniques are used to visualize all accounts and transactions, then complete fraud detection context is available, but the graph becomes unmanageably large and difficult to analyze
Solution Approach 1:
The patent segments the large fraud detection graph into multiple manageable subgraphs based on account relationships and transaction patterns. Each subgraph focuses on a specific account or group of related accounts, allowing analysts to examine detailed context without being overwhelmed by the entire network. This segmentation maintains comprehensive fraud context while improving graph manageability through hierarchical organization.
Solution Approach 2:
The patent extracts relevant subgraphs containing specific fraud detection context from the larger graph structure. By identifying and extracting subgraphs that contain accounts with suspicious activities, relationships, and associated transactions, the system provides focused analysis views that maintain contextual integrity while reducing overall complexity for analyst review.
2Measurement precision
If the graph includes all connected accounts and transactions to provide complete context, then fraud pattern identification is thorough, but navigation and analysis speed decrease
Solution Approach 1:
The patent applies local quality by providing different levels of graph detail based on analytical needs. High-level views show summary statistics and key relationships for quick overview, while detailed views provide complete transaction and account context when needed. This localized adaptation of detail level allows analysts to navigate efficiently at high levels and dive deep into specific areas of interest without constant full-detail rendering.
Solution Approach 2:
The patent implements partial action by dynamically loading and displaying only the necessary portion of the graph data based on current analytical focus. Instead of rendering the entire graph structure simultaneously, the system loads subgraphs and transaction details on-demand as analysts navigate, reducing initial processing time while maintaining access to complete data when required for thorough fraud pattern identification.
3Quantity of substance
If conventional techniques are used to build graphs for fraud detection, then all data is captured, but relevant subgraphs cannot be efficiently identified
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing graph decomposition structures, account relationships, and transaction associations during data ingestion and processing phases. This preliminary organization of data into hierarchical graph structures and pre-calculated relationship indices enables rapid subgraph identification and extraction during analytical operations, significantly improving productivity while maintaining complete data coverage for fraud detection.
Data Source
AI summary
In an embodiment, a process for graph decomposition includes initializing nodes and edges of a data graph for analysis using a computer, and performing message passing between at least a portion of the nodes of the data graph to determine a corresponding measure of interest for each node of at least a portion of the data graph. The process further includes receiving an identification of one or more nodes of interest in the data graph, performing message passing between at least a portion of the nodes of the data graph using at least the determined measures of interest to identify a corresponding subgraph of interest for each of the one or more nodes of interest in the data graph, and performing an analysis action using the one or more identified subgraphs of interest.