Bitwise Transaction Graph Framework for Real-Time Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph technologies face challenges in building real-time scoring models for large datasets, as large volumes of data cannot be quickly loaded into graph database servers, and standard graph learning methods fail to make data separation on a transaction basis, missing real-time modeling requirements.
Innovation Solution
The implementation of a bitwise transaction graph framework that converts historical and current transactions into fixed-length data structures, loading them onto edges of a transaction graph, allowing for real-time fraud detection and decisioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large volumes of data are loaded into graph database servers to build real-time scoring models, then the model accuracy is improved, but the data loading speed decreases and real-time constraints are not met
Solution Approach 1:
The patent segments the graph database into multiple servers, each handling a portion of the data. This allows distributed data storage and processing, enabling the system to handle large volumes of data across multiple servers while maintaining real-time performance through parallel processing and reduced individual server load.
Solution Approach 2:
The patent introduces a new dimension by using a centralized data lake as an additional storage layer. Instead of loading all data into graph database servers, the system stores historical data in a data lake and only loads necessary data into graph servers when needed, adding a temporal and spatial dimension to data management.
2Quantity of substance
If data is partitioned and loaded across many graph database servers to handle large volumes of data, then the data capacity is improved, but the communication overhead increases and real-time constraints are not met
Solution Approach 1:
The patent segments data across multiple graph database servers in a distributed architecture. Each server handles a specific partition of data, enabling the system to scale data capacity while managing communication overhead through efficient data partitioning and selective data retrieval only when needed.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and storing historical data in a centralized data lake before it is needed for real-time modeling. This allows the system to avoid frequent data transfers and communications when building real-time scoring models, as historical data is already prepared and accessible.
3Device complexity
If standard graph learning methods are used to build training and evaluation graphs on data of different time periods, then the data separation is simplified, but the real-time modeling requirements are not met
Solution Approach 1:
The patent introduces dynamic data separation by maintaining a time-aware graph structure that can dynamically separate data based on transaction time. The system uses time windows and temporal filtering to separate training and evaluation data dynamically, enabling real-time modeling while maintaining proper data separation.
Solution Approach 2:
The patent performs preliminary actions by pre-separating and organizing data into time-based partitions in the data lake before real-time modeling. This allows the system to efficiently separate data by time period during real-time processing without adding complex separation logic at runtime, maintaining both simplicity and real-time capability.
Data Source
AI summary
Systems and methods for enabling real-time graph machine learning models using bitwise transaction graph frameworks are disclosed. According to one embodiment, a method may include: (1) receiving, by a bitwise transaction graph computer program, a plurality of historical transactions, wherein each historical transaction comprises a customer identifier for a customer, a card number or card reference number, a merchant identifier for a merchant, a transaction authorization time, a transaction risk score, and a set of real-time fraud risk attributes; (2) converting, by the bitwise transaction graph computer program, the historical transactions to a fixed length data structure; and (3) loading, by the bitwise transaction graph computer program, the fixed length data structure onto edges of a transaction graph, wherein each vertex of the transaction graph represents one of the customers or one of the merchants.


