# US 12,770,155 B2 · Determining more accurate labels for nodes in a graph

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,770,155 B2
- Document ID: US12770155B2
- Status: Granted
- Kind: B2
- Application number: 18/425,153
- Filed: 2024-01-29
- Published: 2025-07-31
- Granted: 2026-10-20
- Earliest claimed: 2024-01-29
- Assignees: Mastercard International Incorporated
- Applicants: Mastercard Technologies Canada Ulc
- Inventors: Rajat Gajendrakumar Patel; Aakarsh Malhotra; Sudipta Modak; Yerramsetty L N Siddhard
- Web page: https://patentgrep.com/patent/US12770155B2

## CPC

- G06Q 30/0185: {Product, service or business identity fraud}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06N 20/20: Ensemble learning
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system for determining more accurate labels for nodes in a graph. The system includes an electronic computing device. The electronic computing device includes an electronic processor. The electronic processor is configured to receive a graph including a plurality of nodes linked by one or more connections. The electronic processor is also configured to augment the graph by creating one or more new connections in the graph. For each node of the plurality of nodes included in the augmented graph, the electronic processor is configured to, using a first machine learning model, determine a first vector associated with the node based on the augmented graph, using a second machine learning model, determine a second vector associated with the node based on the augmented graph, and determine the more accurate label for the node based on the first vector and the second vector.

## Claims

### Claim 1

A system for determining more accurate labels for nodes in a graph, the system comprising:
an electronic computing device, the electronic computing device including:
an electronic processor, the electronic processor is configured to:
train a first machine learning model and a second machine learning model, wherein the first machine learning model and the second machine learning model are initialized with different model weights;
receive a graph including a plurality of nodes linked by one or more connections, wherein
each node of the plurality of nodes represents a card with a chargeback claim or represents a merchant,
each node of the plurality of nodes is unlabeled or associated with a plurality of potential labels, and
a connection between a first node and a second node represents a transaction made between a merchant and a user using a card with a chargeback claim;
augment the graph by creating one or more new connections in the graph;
for each node of the plurality of nodes included in the augmented graph:
using the first machine learning model, determine a first vector associated with the node based on the augmented graph, wherein
each value included in the first vector is associated with a label and represents a likelihood that the label is a more accurate label,
when the node represents a card with a chargeback claim, each label associated with a value included in the first vector is third party fraud, first party fraud, or technical error, and
when the node represents a merchant, each label associated with a value included in the first vector is a merchant category code;
using the second machine learning model, determine a second vector associated with the node based on the augmented graph, wherein
each value included in the second vector is associated with a label and represents a likelihood that the label is the more accurate label,
when the node represents a card with a chargeback claim, each label associated with a value included in the second vector is third party fraud, first party fraud, or technical error, and
when the node represents a merchant, each label associated with a value included in the second vector is a merchant category code; and
determine the more accurate label for the node based on the first vector and the second vector; and
send, to a server, a determination of whether to allow or deny a transaction based on one or more more accurately labeled nodes, wherein the server is configured to perform or deny the transaction based on the determination,
wherein, to train the first machine learning model and the second machine learning model, the electronic processor is further configured to determine a consistency regulation loss value based on the first vector and the second vector.

### Claim 2 (depends on 1)

The system according to claim 1, wherein the electronic processor is configured to train a third machine learning model with training data, wherein the training data includes the labeled nodes.

### Claim 3 (depends on 1)

The system according to claim 1, wherein the electronic processor is configured to augment the graph by creating one or more new connections by:
for each node of the plurality of nodes:
using a fourth machine learning model, predicting a new connection based on a structure of the graph and features associated with the plurality of nodes, wherein the predicted new connection is associated with a likelihood; and
when the predicted new connection is associated with a likelihood above a predetermined threshold, create the new connection in the graph.

### Claim 4 (depends on 3)

The system according to claim 3, wherein the electronic processor is further configured to train the fourth machine learning model by:
determining an augmentation loss value for the fourth machine learning model; and
based on the augmentation loss value, adjusting the fourth machine learning model.

### Claim 5 (depends on 1)

The system according to claim 1, wherein the electronic processor is configured to augment the graph by creating one or more new connections by:
when there are more than a predetermined number of predicted new connections with a likelihood above a predetermined threshold predicted for a node, creating, in the graph, the predetermined number of new connections, wherein the created new connections are predicted new connections that are associated with higher likelihoods.

### Claim 6 (depends on 1)

The system according to claim 1, wherein the electronic processor is configured to determine the more accurate label for the node based on the first vector and the second vector by:
averaging the first vector and the second vector to determine an average vector; and
determining a label associated with a greatest value included in the average vector to be the more accurate label for the node.

### Claim 7 (depends on 6)

The system according to claim 6, wherein the electronic processor is configured to train the first machine learning model and the second machine learning model by:
determining a classifier loss value based on the first vector, the second vector, and a pseudo target vector;
determining a reconstruction loss value based on the average vector and a label corruption matrix; and
updating the first machine learning model and the second machine learning model based on the reconstruction loss value and the classifier loss value.

### Claim 8 (depends on 6)

The system according to claim 6, wherein the electronic processor is configured to train the first machine learning model, the second machine learning model, and a fourth machine learning model by:
determining an augmentation loss value for the fourth machine learning model;
determining a classifier loss value based on the first vector, the second vector, and a pseudo target vector;
determining a reconstruction loss value based on the average vector and a label corruption matrix;
determining a total loss value based on the augmentation loss value, the classifier loss value, and the reconstruction loss value; and
updating the first machine learning model, the second machine learning model, and the fourth machine learning model based on the total loss value.

### Claim 9

A method for determining more accurate labels for nodes in a graph, the method comprising:
training, with an electronic processor, a first machine learning model and a second machine learning model, wherein the first machine learning model and the second machine learning model are initialized with different model weights;
receiving, with the electronic processor, a graph including a plurality of nodes linked by one or more connections, wherein
each node of the plurality of nodes represents a card with a chargeback claim or represents a merchant,
each node of the plurality of nodes is unlabeled or associated with a plurality of potential labels, and
a connection between a first node and a second node represents a transaction made between a merchant and a user using a card with a chargeback claim;
augmenting the graph by creating one or more new connections in the graph;
for each node of the plurality of nodes included in the augmented graph:
using the first machine learning model, determining a first vector associated with the node based on the augmented graph, wherein
each value included in the first vector is associated with a label and represents a likelihood that the label is a more accurate label,
when the node represents a card with a chargeback claim, each label associated with a value included in the first vector is third party fraud, first party fraud, or technical error, and
when the node represents a merchant, each label associated with a value included in the first vector is a merchant category code;
using the second machine learning model, determining a second vector associated with the node based on the augmented graph, wherein
each value included in the second vector is associated with a label and represents a likelihood that the label is the more accurate label,
when the node represents a card with a chargeback claim, each label associated with a value included in the second vector is third party fraud, first party fraud, or technical error, and
when the node represents a merchant, each label associated with a value included in the second vector is a merchant category code; and
determining the more accurate label for the node based on the first vector and the second vector; and
sending, with the electronic processor, a determination of whether to allow or deny a transaction based on one or more more accurately labeled nodes to a server, wherein the server is configured to perform or deny the transaction based on the determination,
wherein training the first machine learning model and the second machine learning model further includes determining a consistency regulation loss value based on the first vector and the second vector.

### Claim 10 (depends on 9)

The method according to claim 9, wherein determining the more accurate label for the node based on the first vector and the second vector includes:
averaging the first vector and the second vector to determine an average vector; and
determining a label associated with a greatest value included in the average vector to be the more accurate label for the node.

### Claim 11

A non-transitory computer-readable medium comprising executable instructions that, when executed by an electronic processor, cause the electronic processor to perform a set of functions comprising:
training a first machine learning model and a second machine learning model, wherein the first machine learning model and the second machine learning model are initialized with different model weights;
receiving a graph including a plurality of nodes linked by one or more connections, wherein
each node of the plurality of nodes represents a card with a chargeback claim or represents a merchant,
each node of the plurality of nodes is unlabeled or associated with a plurality of potential labels, and
a connection between a first node and a second node represents a transaction made between a merchant and a user using a card with a chargeback claim;
augmenting the graph by creating one or more new connections in the graph;
for each node of the plurality of nodes included in the augmented graph:
using the first machine learning model, determining a first vector associated with the node based on the augmented graph, wherein
each value included in the first vector is associated with a label and represents a likelihood that the label is a more accurate label,
when the node represents a card with a chargeback claim, each label associated with a value included in the first vector is third party fraud, first party fraud, or technical error, and
when the node represents a merchant, each label associated with a value included in the first vector is a merchant category code;
using the second machine learning model, determining a second vector associated with the node based on the augmented graph, wherein
each value included in the second vector is associated with a label and represents a likelihood that the label is the more accurate label,
when the node represents a card with a chargeback claim, each label associated with a value included in the second vector is third party fraud, first party fraud, or technical error, and
when the node represents a merchant, each label associated with a value included in the second vector is a merchant category code; and
determining the more accurate label for the node based on the first vector and the second vector; and
sending, to a server, a determination of whether to allow or deny a transaction based on one or more more accurately labeled nodes, wherein the server is configured to perform or deny the transaction based on the determination,
wherein training the first machine learning model and the second machine learning model further includes determining a consistency regulation loss value based on the first vector and the second vector.

### Claim 12 (depends on 11)

The non-transitory computer-readable medium according to claim 11, wherein determining the more accurate label for the node based on the first vector and the second vector includes:
averaging the first vector and the second vector to determine an average vector; and
determining a label associated with a greatest value included in the average vector to be the more accurate label for the node.