Task-Agnostic Embeddings for Payment Entity Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI and ML models in payment ecosystems struggle to learn distinct relationships between entities due to a lack of entity-level semantics, independent representation of entity attributes, and differentiation between static and dynamic attributes in transaction data, leading to inefficiencies in training and deployment.
Innovation Solution
The system generates scalable task-agnostic embeddings by accessing historical transaction data, generating pseudo-objective models for entities, determining entity-specific embeddings, and updating these embeddings when new entities are introduced, allowing for efficient training and redeployment of models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional AI/ML models are trained on transaction data without entity-level semantics, then the model can be trained using standard procedures, but the model fails to learn distinct relationships between different entities (acquirers, merchants, issuers, cardholders)
Solution Approach 1:
The patent segments transaction data into distinct entity-level components by creating separate embedding representations for acquirers, merchants, issuers, and cardholders. Each entity type has its own embedding vector that captures unique characteristics and relationships. This segmentation allows the model to learn distinct relationships between entities while maintaining a structured approach to data organization.
Solution Approach 2:
The patent introduces a new dimensional layer by transforming scalar transaction attributes into vector embeddings. Instead of using traditional scalar features, the system creates multi-dimensional embedding vectors that capture semantic relationships between entities. This dimensional transformation enables the model to learn complex entity relationships that cannot be captured by conventional scalar features.
2Reliability
If task-specific AI/ML models are trained from scratch for each new entity, then the model can achieve task-specific performance, but it leads to the cold start problem with increased processing resources and longer training timelines
Solution Approach 1:
The patent performs preliminary action by pre-training entity-specific embedding models on historical transaction data before deploying task-specific models. The entity embeddings are generated in advance and stored, capturing essential characteristics of acquirers, merchants, issuers, and cardholders. When a new entity is introduced or a task-specific model needs to be trained, these pre-computed embeddings can be directly utilized, eliminating the need to retrain from scratch and significantly reducing training time and computational resources.
Solution Approach 2:
The patent creates reusable entity embedding representations that can be copied and applied across multiple tasks and scenarios. Instead of training separate models for each task, the system generates universal entity embeddings that capture fundamental entity characteristics and can be transferred to different task-specific models. This copying mechanism allows task-specific models to leverage pre-learned entity representations, avoiding redundant training and enabling faster deployment.
3Reliability
If conventional models use aggregated transaction data without differentiating static and dynamic attributes, then the data processing is simpler, but the model cannot understand explicit interactions between static entities (cardholders, issuers) and dynamic entities (merchants, acquirers)
Solution Approach 1:
The patent segments transaction data by explicitly differentiating between static entities (cardholders, issuers) and dynamic entities (merchants, acquirers). Each entity type is assigned its own embedding representation that captures its unique characteristics and interaction patterns. This segmentation enables the model to understand explicit interactions between different entity types while maintaining a structured and organized data representation framework.
Solution Approach 2:
The patent applies local quality by creating entity-specific embedding representations tailored to each entity type's unique characteristics. Instead of using a uniform data representation for all entities, the system generates specialized embeddings for acquirers, merchants, issuers, and cardholders based on their specific attributes and interaction patterns. This localized approach allows the model to capture nuanced entity-specific behaviors and interactions while maintaining overall system coherence.
Data Source
AI summary
Embodiments provide artificial intelligence-based methods and systems for generating scalable task-agnostic embeddings for training task-specific models. Method performed by server system includes accessing historical transaction data from database. Historical transaction data includes entity-specific data, cardholder-specific data, and transaction-specific data. Method includes generating one or more pseudo-objective models for each of plurality of entities based on historical transaction data and one or more pseudo-objectives. Plurality of entities includes acquirer, merchant, and issuer. Method includes determining via one or more pseudo-objective models, entity-specific embeddings for each entity of plurality of entities based on entity-specific data. Entity-specific embeddings includes acquirer-specific embeddings, merchant-specific embeddings, and issuer-specific embeddings. Upon receiving request to induct new entity, method includes accessing new entity data associated with new entity from database. Method includes determining approximate embeddings corresponding to new entity based on new entity data and entity-specific embeddings and updating one of entity-specific embeddings based on approximate embeddings.