Cross-Service Credit Model Transfer for Domain-Specific Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional creditworthiness estimation technologies lack flexibility in enhancing accuracy in specific domains, particularly when transitioning between different financial services.
Innovation Solution
A credit learning device that employs a first training unit to train a model using data from a first service, and a second training unit to adapt this model using data from a second service, with fixed parameters, to estimate credit scores for users across different services.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If transfer learning is applied from credit card loans to small-scale business loans, then the ability to estimate creditworthiness across different domains is improved, but the accuracy for specific domains may deteriorate due to domain differences
Solution Approach 1:
The patent segments users into different user segments based on their service usage patterns and characteristics. By dividing the user base into distinct segments (e.g., based on transaction frequency, amount, or behavior patterns), the system can apply domain-specific models to each segment, thereby maintaining both cross-domain adaptability and segment-specific accuracy.
Solution Approach 2:
The patent applies local quality by training separate models for different user segments or domains while sharing common parameters across models. This allows each domain or segment to have customized model parameters optimized for its specific characteristics, while still benefiting from the transfer learning framework that enables cross-domain application.
2Device complexity
If a single unified model is trained for all services, then device complexity is reduced, but the accuracy of creditworthiness estimation for specific domains deteriorates
Solution Approach 1:
The patent implements universality by creating a multi-functional model system that can handle multiple domains (credit card loans, small-scale business loans, etc.) through a unified transfer learning framework. The system uses a common base model that can be adapted to different domains through parameter sharing and transfer learning, achieving both reduced complexity and maintained accuracy across domains.
Solution Approach 2:
The patent applies parameter changes by training a first model with initial parameters for a first domain, then adapting these parameters to create a second model for a second domain. The system modifies specific parameters while maintaining others, allowing the model to adapt to different domains without requiring complete retraining, thus balancing complexity reduction with domain-specific accuracy.
3Measurement precision
If domain-specific models are trained for each service, then the accuracy of creditworthiness estimation for specific domains is improved, but device complexity and training time increase
Solution Approach 1:
The patent merges the training of multiple domain-specific models into a unified transfer learning process. Instead of training completely separate models for each domain, the system combines them into a single framework where models share common parameters and training resources. This merging approach reduces the overall complexity of model management while maintaining the ability to provide accurate domain-specific estimates.
Solution Approach 2:
The patent applies preliminary action by first training a first model using data from a first domain, then using this pre-trained model as a foundation for training a second model for a different domain. This preliminary training step establishes a base of knowledge that can be transferred and adapted, reducing the need to train models from scratch and thereby reducing overall training time and computational complexity.
Data Source
AI summary
A learning device 1 includes: a first training unit 23a that trains a first model using first training data that includes a combination of a first attribute data group related to a user who uses a first service and a label corresponding to a first score that varies according to risk borne by a service provider when the user uses the first service; and a second training unit 23b that trains a second model, which has some of parameters of a configuration of the trained first model as fixed parameters, using second training data that includes a combination of a second attribute data group related to a user who uses a second service and a label corresponding to a second score that varies according to risk borne by a service provider when the user uses the second service.