RNN Profile Extraction for Real-Time Multi-Entity Fraud Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for determining entity profiles in transactional data require manual engineering, extensive computational resources, and are inefficient for real-time processing, especially when dealing with multiple entities, leading to incomplete context and suboptimal fraud detection performance.
Innovation Solution
Utilizing parallel trained recurrent neural networks (RNNs) to learn entity-specific embeddings in real-time, which are then combined by a classifier to provide accurate multi-entity context for fraud detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual engineering of profile features is used, then context accuracy is improved, but device complexity and resource requirements increase
Solution Approach 1:
The system uses automatic profile extraction where the RNN models automatically learn and extract profile features from transactional data streams without manual intervention. The models self-adapt to entity behavior patterns, eliminating the need for manual feature engineering while maintaining high context accuracy.
Solution Approach 2:
The patent replaces manual mechanical feature engineering with automated neural network-based feature extraction. The RNN models automatically discover and extract relevant profile features from raw data, substituting the manual process with an intelligent automated system that reduces complexity.
2Loss of information
If traditional profile methods are used, then entity context is captured, but processing speed and real-time performance deteriorate
Solution Approach 1:
The system pre-trains multiple RNN models offline on historical transactional data to learn entity behavior patterns. During real-time processing, these pre-trained models quickly extract profiles from incoming data streams without requiring complex runtime computations, enabling both complete context capture and high processing speed.
Solution Approach 2:
The patent divides the profile extraction task into separate specialized RNN models for different entity types (e.g., customer profiles, merchant profiles, device profiles). Each model is optimized for its specific entity type, allowing parallel processing and improving overall system throughput while maintaining comprehensive context.
3Reliability
If multiple entity profiles are maintained, then detection accuracy is improved, but storage capacity and computational resources increase
Solution Approach 1:
The system transforms profile data into compact vector representations using the RNN models. Instead of storing extensive raw profile information, the models compress entity behavior patterns into efficient vector forms that occupy minimal storage space while preserving the essential characteristics needed for accurate fraud detection.
Solution Approach 2:
The patent uses embedding vectors as compressed copies of entity profiles. These vector representations capture the essential behavior patterns of entities in a space-efficient manner, allowing the system to maintain multiple entity profiles with minimal storage requirements while preserving detection accuracy.
4Stability of the object's composition
If state management is implemented for profiles, then context tracking is improved, but system complexity and resource consumption increase
Solution Approach 1:
The RNN models automatically maintain and update entity profiles as new transactional data arrives in the stream. The models self-adapt to changing entity behavior patterns without requiring external state management intervention, simplifying the system while maintaining accurate context tracking.
Solution Approach 2:
The system continuously updates entity profiles as new transactions arrive, with the RNN models processing data streams in real-time. This continuous learning process maintains up-to-date context information without requiring periodic batch processing or complex state management interventions.
Data Source
AI summary
In various embodiments, a process for automatic profile extraction in data streams using recurrent neural networks includes receiving input sequence data associated with a stream of events and using a plurality of trained recurrent neural network machine learning models at least in part in parallel to determine different embedding output sets that represent at least a portion of the input sequence data in a plurality of different embedding spaces. The process includes providing the different embedding output sets to one or more classifier machine learning models to determine one or more classifier results, and using the one or more classifier results to provide a prediction output.