User Embedding Mapping for Adaptive Access Score Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current prediction models for access requests are not accurate in determining the authenticity of user behavior changes, particularly when authorized users shift contexts, leading to potential fraudulent transactions.
Innovation Solution
A system that generates access scores using user embeddings and transformation models to incorporate updated contextual information, allowing for continuous adaptation to user behavior changes without frequent retraining of the prediction model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the prediction model is frequently retrained to incorporate user behavior changes, then the accuracy of access score determination is improved, but the computational resources and time required increase
Solution Approach 1:
The system segments the user behavior representation into two independent components: a static user embedding vector that captures general user characteristics and remains unchanged, and a dynamic contextual embedding that captures recent behavior patterns and is updated frequently. This segmentation allows the system to incorporate updates efficiently without retraining the entire model.
Solution Approach 2:
The patent introduces an intermediary transformation model that maps between different embedding spaces. When user context changes, the system uses this transformation model to convert new contextual embeddings into the existing embedding space without requiring retraining of the main prediction model, thus resolving the contradiction between accuracy and computational time.
2Measurement precision
If the prediction model incorporates updated user context information, then the accuracy reflects current user behavior, but the computational complexity increases
Solution Approach 1:
The system separates the user representation into a static component (user embedding) and a dynamic component (contextual embedding). The static component maintains computational simplicity by remaining unchanged, while the dynamic component handles updates efficiently through embedding table lookups rather than complex retraining operations.
Solution Approach 2:
Instead of changing the fundamental model structure or retraining parameters, the system updates only the embedding table parameters which store contextual information. This parameter change approach maintains low computational complexity while achieving accurate reflection of current user behavior.
3Measurement precision
If the prediction model uses detailed contextual information about user behavior, then the accuracy of fraud detection improves, but the amount of data processing required increases
Solution Approach 1:
The system extracts only the most relevant contextual information needed for fraud detection and represents it as a condensed embedding vector. By taking out and compressing essential behavioral patterns into a fixed-dimensional representation, the system achieves high detection accuracy while minimizing data processing requirements.
Solution Approach 2:
The patent transforms detailed contextual data into a lower-dimensional embedding space. Instead of processing high-dimensional raw data, the system works with compressed embedding vectors that capture the essential information in a more efficient dimensional representation, thereby improving processing efficiency without sacrificing detection accuracy.
Data Source
AI summary
A method and system for determining an access score is disclosed. The method includes receiving an access request to access a resource by a user device. Next, a user embedding is retrieved from an embedding table, the user embedding associated with a user identifier of the user device and providing a multidimensional data point that represents a context of a user identifier. The context may correspond to the user identifier appearing in previous access requests within temporal proximity to other access requests from a subset of other user devices among a plurality of user devices. The method then inputs the user embedding into a first machine learning model that is trained based at least in part on the embedding table. The first machine learning model subsequently outputs an access score that corresponds to a level of authenticity of authorizing the user device to access the resource.