Entity Graph Risk Calculation Using Edge Weights by Cardinality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing risk calculation methods for electronic transactions are computationally expensive and not well-suited for real-time assessment, limiting their effectiveness in fraud detection and transaction processing.
Innovation Solution
A message passing framework that incorporates edge weights based on inverse cardinality and learned weight matrices, allowing for the propagation of information across connected nodes in an entity graph, which can be processed by traditional machine learning algorithms, reducing the need for complex neural network architectures and enhancing computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If graph neural networks are used to incorporate information from connected nodes for risk prediction, then fraud detection accuracy is improved, but computational cost increases and real-time assessment becomes difficult
Solution Approach 1:
The patent segments the complex graph neural network computation into two distinct phases: an offline training phase where the graph structure and node features are pre-processed and stored, and an online inference phase where only simple aggregation operations are performed. This segmentation allows the computationally intensive work to be done beforehand, enabling real-time risk assessment during transaction processing without sacrificing fraud detection accuracy.
Solution Approach 2:
The patent performs preliminary computation by pre-calculating and storing graph structures, node features, and relationship information during an offline training phase. This preliminary action includes building the entity graph, computing node embeddings, and storing aggregated features that can be quickly retrieved during online transaction risk assessment, eliminating the need for complex real-time graph computations.
2Measurement precision
If complex neural network architectures are used to capture spatial dependencies and differentiate relationships, then risk calculation accuracy is improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent replaces complex, expensive neural network architectures with simpler, computationally inexpensive aggregation operations that use pre-computed graph features. Instead of deploying heavy deep learning models for real-time inference, the system uses lightweight operations that aggregate pre-processed node features according to the graph structure, achieving comparable accuracy with minimal computational resources and system complexity.
Solution Approach 2:
The patent creates simplified copies of the graph structure and node features in a pre-processed format that can be easily queried during online inference. Rather than maintaining and querying the full complex graph neural network model in real-time, the system uses copied and pre-aggregated feature representations that preserve the essential spatial dependencies and relationship information while enabling fast, simple computations.
Data Source
AI summary
A transaction processing method includes receiving data respective of an electronic transaction, the data including an account involved in the transaction and determining an entity graph according to the account. The entity graph includes a plurality of first-order connections, each including a primary node representative of the account, a secondary node representative of a secondary entity, and an edge from the primary node to the secondary node, and a plurality of second-order connections, each including one of the secondary nodes, a tertiary node representative of a tertiary entity, and an edge from the secondary node to the tertiary node. The method includes determining a weight associated with each first-order connection edge based on a quantity of second-order connections associated with the secondary node in the first-order connection, calculating a risk associated with the electronic transaction according to the respective weights, and processing the electronic transaction according to the calculated risk.