Data Record Matching Model Fine-Tuning for Customer-Specific Accuracy

Solution Overview

Problem

Conventional data record matching models trained with generic data exhibit sub-optimal accuracy when applied to customer data due to lack of customization.

Innovation Solution

A two-phase learning process is employed to adjust parameters of the matching model using customer-specific labeled data, fine-tuning feature vector coefficients to enhance matching accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a data record matching model is trained using generic computer-generated training data, then the model can be quickly deployed with general matching capabilities, but the matching accuracy deteriorates when applied to specific customer data

Engineering Contradiction:

Improvemodel deployment speedVSAvoidmatching accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary training using generic computer-generated data to establish baseline matching capabilities, then applies fine-tuning using customer-specific labeled data to adapt the model parameters. This two-stage approach allows the model to be quickly deployed with general capabilities while subsequently improving accuracy for specific customer applications through parameter adjustment.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention adjusts the parameters of the pre-trained matching model using customer-specific labeled data. By changing the model parameters through fine-tuning with customer data, the system transforms a generic model into a customized one that achieves higher matching accuracy for specific customer applications without retraining the entire model from scratch.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If a data record matching model is customized for specific customer data, then matching accuracy improves, but the complexity of model training and tuning increases

Engineering Contradiction:

Improvematching accuracyVSAvoidmodel training complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary training using generic computer-generated data to establish baseline matching capabilities, then applies fine-tuning using customer-specific labeled data to adapt the model parameters. This two-stage approach allows the model to be quickly deployed with general capabilities while subsequently improving accuracy for specific customer applications through parameter adjustment.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention uses customer-specific labeled data as a copy or supplement to the generic training data. Instead of creating entirely new training data from scratch, the system leverages existing generic training data and augments it with customer-specific labeled examples, reducing the overall complexity of model customization while maintaining high accuracy.

Inventive Principle:

Principle #26Copying

3Ease of manufacture

If conventional matching models use computer-generated training data, then training data can be readily obtained without customer involvement, but the model lacks tailoring to customer data characteristics

Engineering Contradiction:

Improvetraining data availabilityVSAvoidcustomer data tailoring

Core Design Contradiction:

Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary training using generic computer-generated data to establish baseline matching capabilities, then applies fine-tuning using customer-specific labeled data to adapt the model parameters. This two-stage approach allows the model to be quickly deployed with general capabilities while subsequently improving accuracy for specific customer applications through parameter adjustment.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention creates a multi-functional training approach where the same matching model framework can be applied to multiple customers by combining generic pre-training with customer-specific fine-tuning. This universal approach allows the system to serve different customers with varying data characteristics without requiring completely separate models for each customer.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12608652B2Tuning a trained data record matching model using customer data and representation learning

Publication Date: 2026.04.21 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

A method, system, and computer program product are configured to create a tuned data record matching model by adjusting values of one or more parameters in a data record matching model based on a second training data set labeled at a data record level, wherein the data record matching model is initially trained using a first training data set labeled at an attribute level.