Mini-Graph Fraud Detection Using Transaction Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number of electronic and remote transactions poses challenges in authenticating the identity of parties involved, as fraudsters employ sophisticated methods such as photo editing and facial morphing to deceive entities, leading to potential identity theft and other security breaches.
Innovation Solution
A mini-graph based fraud detection system that generates embeddings from transaction data, including image snippets and text, to create a mini-graph for classification using machine learning, enabling efficient detection of fraudulent activities by analyzing similarities with prior transactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional identity verification methods are used, then the process is simple, but fraudsters can easily deceive using photo editing and facial morphing
Solution Approach 1:
The patent segments the verification process into multiple independent components: image processing module, embedding generation module, graph construction module, and classification module. Each component handles a specific aspect of verification, allowing the system to detect sophisticated fraud while maintaining modular simplicity for implementation and maintenance.
Solution Approach 2:
The patent introduces an intermediary embedding representation that bridges the gap between raw image data and fraud detection classification. The embedding layer transforms complex image information into a compressed representation that captures essential features, enabling accurate fraud detection without requiring the entire complex verification system to be overly complicated.
2Reliability
If comprehensive fraud detection analysis is performed, then detection accuracy improves, but processing time and computational cost increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing embeddings for transaction images before the actual fraud detection is needed. The graph structure and embedding generation are prepared in advance, allowing rapid query processing when transactions occur, thus reducing real-time processing time while maintaining high detection accuracy.
Solution Approach 2:
The patent extracts only the most critical information from the full transaction data by generating compressed embeddings that capture essential fraud indicators. This extraction process filters out redundant information while preserving the most important features for fraud detection, enabling fast processing without sacrificing accuracy.
3Measurement precision
If detailed image analysis is performed, then detection precision improves, but system resource consumption increases
Solution Approach 1:
The patent changes the parameter representation from raw pixel data to compressed embedding vectors. This parameter transformation reduces the amount of data that needs to be processed while maintaining the essential information needed for precise fraud detection, thereby reducing computational resource consumption without sacrificing analysis precision.
Solution Approach 2:
The patent creates a simplified copy of the image data in the form of embeddings that retain the essential fraud detection capabilities. This copy allows the system to perform detailed analysis on a compressed representation rather than the full original data, reducing resource consumption while maintaining detection precision.
Data Source
AI summary
The disclosure includes a system and method for obtaining transaction data associated with a transaction under test including image data associated with a first image; generating a set of transaction parts including a first transaction part and a second transaction part; generating a first embedding; obtaining a first result set based on a first transaction part comprising a first set of embeddings associated with a first set of prior transactions; obtaining a second result set based on a second transaction part associated with a second set of prior transactions; generating a mini-graph associated with the transaction under test; and classifying, using a first machine learning model, whether the mini-graph associated with the transaction under test is indicative of fraud.