Distributed User Model Adaptation for Fraud Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for monitoring user authenticity during activities on application servers are inadequate in accurately distinguishing genuine users from fraudulent ones, as they rely on limited behavioral analysis and do not adapt effectively to changing user patterns over time.

Innovation Solution

A distributed server system comprising an application server and a user-model server that performs a user-modelling process to adapt user models session-by-session, using data mining algorithms like hidden Markov models and neural networks to analyze user activity data, and triggers corrective actions based on a total risk-score value calculated from feature-specific risk-score values.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional rule-based behavioral analysis is used to monitor user authenticity, then the system is simple to implement and operate, but the accuracy of distinguishing genuine users from fraudulent ones is insufficient

Engineering Contradiction:
Improveuser authenticity detection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments user behavior monitoring into multiple independent feature dimensions (login time, location, device, browsing pattern, transaction behavior). Each dimension is analyzed separately by dedicated modules, and their results are aggregated to form a comprehensive authenticity assessment. This segmentation enables precise multi-faceted analysis while maintaining manageable system complexity through modular design.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system implements dynamic user profiling that continuously adapts to changing user behavior patterns over time. Instead of static rules, the system learns and updates user-specific behavioral baselines, allowing it to accurately detect deviations that indicate fraud while adapting to legitimate changes in user habits. This dynamic approach significantly improves detection accuracy compared to fixed rule-based systems.

Inventive Principle:
Principle #15Dynamics

2Adaptability or versatility

If static user behavior rules are applied for fraud detection, then the system is easy to operate, but it cannot adapt effectively to changing user patterns over time

Engineering Contradiction:
Improveadaptation to changing user patternsVSAvoidsystem operation simplicity
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The system incorporates continuous feedback loops where detection results and user responses are fed back into the learning modules. This feedback mechanism allows the system to automatically refine its understanding of legitimate versus fraudulent behavior patterns. The feedback-driven adaptation enables the system to stay current with evolving fraud techniques and legitimate user behavior changes without requiring manual rule updates, maintaining ease of operation while achieving high adaptability.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system performs preliminary learning and baseline establishment during normal user activity periods before fraud detection becomes critical. By pre-adapting to seasonal patterns, typical user behavior changes, and emerging fraud trends in advance, the system is already optimized when actual fraud detection is needed. This preliminary adaptation reduces the operational burden during critical detection phases while maintaining high adaptability.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If comprehensive multi-feature behavioral analysis is performed, then the detection accuracy of fraudulent activities is improved, but the processing time and computational resources increase

Engineering Contradiction:
Improvefraudulent activity detection accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system implements a tiered analysis approach where not all behavioral features are evaluated with equal depth for every user interaction. High-risk indicators trigger more comprehensive analysis, while low-risk routine operations receive streamlined processing. This partial action strategy maintains high detection accuracy for critical cases while reducing average processing time through selective deep analysis only when necessary.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system pre-computes and caches baseline behavioral profiles and normal pattern templates during off-peak periods. When real-time fraud detection is needed, the system compares current behavior against these pre-prepared references rather than computing everything from scratch. This preliminary preparation dramatically reduces real-time processing requirements while maintaining comprehensive analysis accuracy for fraud detection.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3203436B1Monitoring user authenticity in distributed system
Publication Date: 2022.05.18 AMADEUS SAS
  • EP3203436B1 patent drawingFigure 1
  • EP3203436B1 patent drawingFigure 2
  • EP3203436B1 patent drawingFigure 3a

AI summary

A method of monitoring user authenticity during user activities in a user session on at least one application server is provided. The method being carried out in a distributed manner by means of a distributed server system. The method comprises a user modelling-process and a user-verification process. The user-modelling process is performed on a user-model server, wherein an existing user model is adapted session-by-session to user-activity data received from the application server. The user-verification process is performed on the at least one application server on the basis of the user model adapted on the user-model server. The user-verification process comprises comparing the user model with features extracted from user activity in the user session on the application server and determining a total risk-score value on the basis of the comparison. In response to the total risk-score value exceeding a given threshold, a corrective action is performed.