Credit Scoring for Behavior-Unknown Users via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing credit scoring models are unable to accurately assess behavior-unknown users, limiting the coverage of credit-related services and resulting in a low user conversion rate on application platforms.
Innovation Solution
A method and server configuration that identifies target user sets with higher predicted profit values among behavior-unknown users, determines their credit scores based on behavior-known users, selects user samples with satisfactory profit values, and trains a scoring model to perform accurate credit scoring for behavior-unknown users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If credit scoring models are trained only on behavior-known users, then the scoring accuracy for behavior-known users is improved, but the coverage of credit-related services remains limited and user conversion rate is low
Solution Approach 1:
The patent segments users into behavior-known users and behavior-unknown users, and further divides behavior-unknown users into different groups (first behavior-unknown users with higher predicted profit values and second behavior-unknown users). This segmentation allows the system to apply different scoring approaches to different user segments, thereby expanding service coverage while maintaining scoring accuracy for each segment.
Solution Approach 2:
The patent introduces an intermediary approach by using behavior-known users as a reference group to establish scoring models that can then be applied to behavior-unknown users. The system uses the credit scores and behavior patterns of behavior-known users as intermediaries to infer and predict credit scores for behavior-unknown users, enabling service expansion without direct behavioral data from the target users.
2Reliability
If user samples are selected only from behavior-known users, then the reliability of training data is improved, but the quantity of applicable users for credit services is reduced
Solution Approach 1:
The patent applies preliminary action by first determining credit scores for behavior-unknown users using predictive methods before actually using these users as training samples. The system pre-assesses the creditworthiness of behavior-unknown users based on limited available data and predicted profit values, then selects those with higher scores as training samples, thereby ensuring data reliability while expanding the pool of eligible users.
Solution Approach 2:
The patent changes the selection parameters for training samples by incorporating predicted profit values as an additional criterion beyond traditional behavior data. By adjusting the selection parameters to include behavior-unknown users with high predicted profit values, the system expands the quantity of training samples while maintaining reliability through the profit value filter.
Data Source
AI summary
A credit scoring method and a server are provided. The method includes: determining at least one target user set in a plurality of user sets. The at least one target user set includes first behavior-unknown users whose predicted profit values are higher than second behavior-unknown users, the second behavior-unknown users are users other than the first behavior-unknown users on an application platform. The method also includes determining credit scores of first behavior-unknown users according to credit scores of behavior-known users in at least one determined target user set; obtaining first user samples from the first behavior-unknown users according to credit records of the first behavior-unknown users, the first user samples having profit values that satisfy a preset threshold; and performing credit scoring for a user according to a second scoring model obtained through training a first scoring model according to the first user samples and second user samples.