Merchant Embedding Refinement for Transaction Store Name Merging

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Merchant stores are often identified inaccurately in transaction data, leading to confusion and incorrect associations with different merchants, which complicates transaction analysis and customer understanding.

Innovation Solution

A system utilizing pre-trained word embeddings refined by machine-learning models to identify and aggregate merchant stores by fine-tuning embeddings, clustering similar merchant names, and normalizing them to accurately merge clusters based on similarity and anchor names.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If merchant stores are identified using different identifiers in transaction data, then the system can capture diverse merchant representations, but the merchant stores are incorrectly identified as stores associated with different merchants

Engineering Contradiction:
Improvemerchant identifier diversityVSAvoidmerchant identification accuracy
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent merges multiple merchant identifier clusters into a single unified merchant entity by computing similarity between cluster names and merging clusters that represent the same merchant. This resolves the contradiction by combining diverse identifiers (improving adaptability) while maintaining accurate merchant identification (preserving reliability) through the merging process that consolidates variations of the same merchant name.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces cluster names as intermediary representations that bridge raw merchant identifiers and final merchant entities. By computing similarity between these intermediary cluster names and comparing them against known merchant names, the system can accurately identify and merge clusters representing the same merchant, thus resolving the identification accuracy issue while preserving identifier diversity.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If machine-learning models are used to refine word embeddings, then the accuracy of merchant data aggregation is improved, but the computational complexity and processing time increase

Engineering Contradiction:
Improvemerchant data aggregation accuracyVSAvoidsystem computational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary clustering of merchant embeddings into named clusters before the final matching process. By pre-organizing embeddings into meaningful clusters with representative names, the system reduces the computational complexity of the subsequent matching stage, while still achieving high accuracy through the refined cluster-level comparisons rather than raw embedding comparisons.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent extracts cluster names from sets of merchant embeddings as representative identifiers. By taking out the essential naming information from complex embedding data, the system simplifies subsequent processing while preserving the accuracy needed for merchant identification, thus reducing computational complexity without sacrificing measurement precision.

Inventive Principle:
Principle #2Taking out (Extraction)

3Adaptability or versatility

If multiple merchant identifier clusters are created, then the system can capture variations in merchant naming, but confusion in identifying parties to a transaction increases

Engineering Contradiction:
Improvemerchant naming variation captureVSAvoidtransaction party identification clarity
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent implements a feedback mechanism by comparing cluster names against known merchant names and using this comparison to determine whether to merge clusters. This feedback loop resolves confusion by systematically evaluating whether multiple clusters represent the same merchant, thus maintaining adaptability to naming variations while restoring clarity to transaction party identification through informed merging decisions.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4647998A1System and method for leveraging pre-trained embeddings to detect and merge merchant data
Publication Date: 2025.11.12 FISERV INC
  • EP4647998A1 patent drawingFigure 1
  • EP4647998A1 patent drawingFigure 2
  • EP4647998A1 patent drawingFigure 3

AI summary

A system including one or more processors and a computer-readable, non-transitory medium including instructions which cause at least one of the one or more processors to obtain merchant data including a plurality of merchants, obtain a set of word embeddings extracted using a large language model, refine the set of word embeddings by executing a machine-learning model using as input the merchant data to obtain a set of merchant embeddings, determine a first cluster of first merchant embeddings and a second cluster of second merchant embeddings within the set of merchant embeddings, determine a first name for the first cluster based on the first embeddings and a second name for the second cluster based on the second embeddings, and merge the first cluster and the second cluster based on a similarity of the first name and the second name to obtain a merged cluster.