Fraud Detection Model Using Multipartite Graph Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer systems face challenges in detecting and mitigating fraudulent attempts to access secure electronic resources, particularly when malicious actors compromise legitimate user accounts, leading to unauthorized transactions and potential financial or reputational harm.
Innovation Solution
A fraud detection model using a multipartite graph embedding approach is implemented, which represents sender and recipient accounts, as well as IP addresses, to evaluate requests and predict fraudulent activity by updating embedding values based on past transactions and behaviors, enabling real-time detection and prevention of fraudulent access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fraud detection methods are used, then the system structure remains simple, but the detection accuracy and ability to identify compromised accounts is insufficient
Solution Approach 1:
The patent segments the fraud detection system into multiple independent components: a graph construction module that builds the multipartite graph from transaction data, an embedding generation module that creates node representations, and a fraud detection module that evaluates requests. This segmentation allows each component to be optimized independently while working together to achieve high detection accuracy through the integrated graph embedding approach.
Solution Approach 2:
The patent transforms traditional flat transaction data into a multi-dimensional graph structure where nodes represent entities (accounts, devices, locations) and edges represent relationships. The multipartite graph embedding then projects this complex graph into a lower-dimensional vector space, capturing intricate fraud patterns that would be invisible in traditional two-dimensional transaction tables, thereby significantly improving detection precision.
2Speed
If real-time fraud detection is implemented, then the response time to fraudulent requests is reduced, but the computational resources and processing complexity increase
Solution Approach 1:
The system performs preliminary actions by pre-constructing the multipartite graph from historical transaction data and pre-computing node embeddings during off-peak times. When a fraud detection request arrives, the system only needs to evaluate the pre-built graph structure and compare it against the current request, rather than building the entire analysis from scratch in real-time. This preliminary preparation enables fast real-time responses while keeping the computational burden manageable.
Solution Approach 2:
The patent creates a simplified representation (embedding vector) of each node in the complex graph structure. These embedding vectors capture the essential characteristics and relationships of accounts, devices, and locations in a compressed format. When evaluating a fraud request, the system works with these lightweight vector copies rather than the full complex graph data, dramatically reducing the computational complexity of real-time detection while maintaining detection accuracy.
3Adaptability or versatility
If the fraud detection model updates embedding values continuously, then the adaptability to changing fraud patterns improves, but the data processing load and model maintenance complexity increase
Solution Approach 1:
The patent implements a feedback mechanism where the fraud detection system continuously monitors new transactions and updates the graph embedding model based on observed patterns. When fraudulent behavior is detected or new fraud patterns emerge, the system uses this feedback to retrain and update the embedding vectors, allowing it to adapt to changing fraud tactics. This feedback loop ensures the model remains current and effective against evolving fraud strategies while maintaining a manageable update process through incremental learning techniques.
Data Source
AI summary
A fraud detection model is used by a computer system to evaluate whether to grant a request to access a secure electronic resource. Before granting the request, the computer system evaluates the request using a multi-partite graph model generated using a plurality of previous requests. The multi-partite graph model includes at least a first set of nodes for sender accounts, a second set of nodes for recipient accounts, and a third set of nodes.