Byte Pair Encoding for Fast Fraud Screening of Account Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection methods in computer-based electronic networks are time-consuming, data-intensive, and ineffective in identifying fraudulent account creation, particularly due to the rise of AI bots, leading to significant resource expenditure and security breaches.
Innovation Solution
Implementing a fraud detection system that uses byte pair encoding to generate a data compression vocabulary from legitimate account data, compressing new account data, and comparing its length to a threshold to determine fraudulent activity, thereby preventing unauthorized account creation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional fraud detection methods are used to analyze account data, then comprehensive security checking is achieved, but the process becomes time-consuming and data-intensive
Solution Approach 1:
The patent extracts only the most critical detection feature (data length after compression) from the complete account data, eliminating the need to analyze all data elements. This extraction approach maintains fraud detection capability while dramatically reducing processing time and computational resources required.
Solution Approach 2:
The patent segments the account data processing into distinct phases: compression using byte pair encoding, length calculation, and threshold comparison. This segmentation allows the system to focus only on the essential detection aspect (length measurement) rather than进行全面 data analysis, resolving the contradiction between thorough detection and processing efficiency.
2Reliability
If comprehensive account data analysis is performed to detect fraud, then detection accuracy is improved, but resource expenditure increases significantly
Solution Approach 1:
The patent extracts only the essential detection signal (data length) from the complete account data using byte pair encoding compression. This extraction eliminates the need for resource-intensive comprehensive analysis while preserving fraud detection accuracy, directly resolving the contradiction between detection reliability and resource consumption.
Solution Approach 2:
The patent uses a lightweight, computationally inexpensive compression algorithm (byte pair encoding) that requires minimal processing resources compared to traditional fraud detection methods. This disposable-like approach to data transformation enables rapid processing with negligible resource expenditure while maintaining effective fraud detection.
3Reliability
If traditional fraud detection systems process large volumes of account data, then thorough security screening is achieved, but data storage requirements increase
Solution Approach 1:
The patent extracts only the critical security indicator (compressed data length) from the complete account data, storing only this essential metric rather than the full data set. This extraction maintains security screening effectiveness while minimizing data storage requirements to the absolute minimum necessary for fraud detection.
Solution Approach 2:
Instead of storing and analyzing complete account data to detect fraud, the patent inverts the approach by compressing the data and storing only the length metric. This inversion fundamentally reduces storage requirements from gigabytes to bytes per record while preserving fraud detection capability through the length-based analysis.
Data Source
AI summary
Techniques are described herein for fraud detection which may include generating, from legitimate account data instances, a data compression vocabulary. The data compression vocabulary may specify one or more tokens identified from the legitimate account data instances and a corresponding replacement value. A new instance of account data may be received. Compressed account data may be generated based at least in part on replacing a token identified in the account data with a respective replacement value identified for the token in the data compression vocabulary. Any amount by which the new account data was compressed may be computed and used to make a determination as to whether the new instance of account data is fraudulent or legitimate.