Graph Embedding for Abnormal Order Payment Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional methods for detecting abnormal financial transactions in e-commerce platforms are rigid, prone to false alarms, time-consuming, and struggle with the complexity and volume of data, hindering scalability and timely response to discrepancies.

Innovation Solution

Utilizing a deep graph learning method that constructs a graph from transaction data, applies graph embedding techniques, and employs anomaly detection to identify abnormal payment behavior, supported by a graph visualization tool for transparent verification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional rule-based methods are used for transaction monitoring, then the system is simple to implement, but it generates false alarms and misses complex anomalies

Engineering Contradiction:
Improveaccuracy of anomaly detectionVSAvoidcomplexity of detection system
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces traditional rule-based mechanical detection systems with a machine learning-based graph neural network system. The GNN model learns complex patterns from transaction data represented as graphs, where nodes are accounts and edges are transactions. This substitution enables the system to detect complex anomalies that rule-based systems miss, while the automated learning process reduces false alarms compared to rigid rule-based approaches.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent employs a composite approach by combining multiple techniques: graph neural networks for pattern recognition, anomaly detection algorithms for identifying deviations, and visualization tools for interpretation. This composite system integrates the strengths of different methods to achieve high reliability in detecting complex financial anomalies while managing system complexity through modular architecture.

Inventive Principle:
Principle #40Composite materials

2Productivity

If manual reconciliation methods are used, then the process is transparent and verifiable, but it is time-consuming and creates bottlenecks

Engineering Contradiction:
Improvespeed of transaction processingVSAvoidtime for reconciliation process
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system implements self-service through automated anomaly detection and reconciliation processes. The graph neural network automatically analyzes transaction patterns, identifies anomalies, and generates reconciliation reports without human intervention. The visualization tools provide self-explanatory graphical representations that enable stakeholders to verify results independently, eliminating the need for time-consuming manual reconciliation while maintaining transparency.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces manual mechanical reconciliation processes with automated computational systems. The machine learning model processes transactions at scale, identifying patterns and anomalies far faster than manual methods. The system substitutes human analysts with automated algorithms that continuously monitor transactions, significantly reducing processing time and eliminating bottlenecks while maintaining auditability through visualization tools.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If traditional methods are used to handle large volumes of transaction data, then the system is easy to operate, but it struggles with data volume and complexity

Engineering Contradiction:
Improveability to handle complex financial ecosystemsVSAvoidcomplexity of data processing system
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent transforms transaction data into a graph structure where accounts are nodes and transactions are edges, adding a relational dimension to the data. This graph representation captures complex relationships and patterns that traditional flat data structures cannot represent. The graph neural network then processes this multi-dimensional structure, enabling the system to adapt to intricate financial ecosystems while managing complexity through the structured graph framework.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The graph neural network system is designed to be universal and adaptable to various financial ecosystems. The same GNN architecture can process different types of transaction data, account relationships, and anomaly patterns across diverse e-commerce platforms. This multi-functional approach enables the system to handle complex financial ecosystems without requiring separate specialized systems for each scenario.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Adaptability or versatility

If rule-based detection systems are used, then the system is fast to implement, but it is rigid and cannot adapt to rapid business growth

Engineering Contradiction:
Improveadaptability to business growthVSAvoidtime to update detection rules
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent implements a dynamic system where the graph neural network continuously learns from new transaction data, automatically adapting to changing business patterns and growth. Unlike static rule-based systems that require manual updates, the GNN model dynamically adjusts its parameters and patterns based on incoming data, enabling the system to keep pace with rapid business growth without requiring time-consuming rule updates.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system incorporates feedback mechanisms where detection results and new transaction data continuously feed back into the training process. The graph neural network learns from both normal and anomalous patterns, continuously improving its detection capabilities. This feedback loop enables automatic adaptation to business growth and changing transaction patterns, eliminating the need for manual rule updates while maintaining high adaptability.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4715714A1System and method for detecting abnormal order payment behaviour using graph model embedding and anomaly detection
Publication Date: 2026.03.25 EBAY INC
  • EP4715714A1 patent drawingFigure 1
  • EP4715714A1 patent drawingFigure 2A~2B
  • EP4715714A1 patent drawingFigure 3

AI summary

Some aspects of the present technology relate to technologies for detecting abnormal payment behavior using graph model embedding and anomaly detection. In accordance with some configurations, order payment data is collected from various sources, including e-commerce platforms, financial institutions, and payment processors. The collected payment data is structured as a graph for each order. Nodes represent individual payment transactions related to the order. Graph embedding techniques are applied to transform the payment data graph into a numerical vector space representation. The embedded data is analyzed for a particular interval of time to identify recurring patterns. A baseline for normal patterns is established for the interval of time and any patterns that deviate significantly from the baseline are flagged as potential abnormal payment behaviors. In some aspects, a graph visualization comparison tool aids in the transparent verification of reconciliations and provides intuitive insights for stakeholders.