# US 2026/0087034 A1 · Computer-based systems configured for instance dependent optimization of information processing systems and methods of use thereof

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

- Number: US 2026/0087034 A1
- Document ID: US20260087034A1
- Status: Published application
- Kind: A1
- Application number: 19/405,051
- Filed: 2025-12-01
- Published: 2026-03-26
- Earliest claimed: 2024-03-13
- Assignees: Capital One Services, LLC
- Applicants: Capital One Services, LLC
- Inventors: Ali S. Al-Shehab; Jr-Wei Jeng; Niti N. Sheth; Tanveer Afzal Faruquie; David Edward Lutz; Nathan L. Sheridan
- Web page: https://patentgrep.com/patent/US20260087034A1

## CPC

- G06F 16/287: {Visualization; Browsing}
- G06F 16/2237: {Vectors, bitmaps or matrices}
- G06F 40/295: Named entity recognition
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

In order to facilitate entity resolution, systems and methods include a processor receiving a plurality of records associated with one or more entity records. The processor utilizes a first natural language processing model to determine a set of clusters. The processor then utilizes instance inputs to determine adjustments to the natural language processing model to determine from the groups of clusters a second map of clusters from the entity feature, then determines a merge of the entity records, and displays the merged entity records.

## Claims

### Claim 1

A computer-implemented method comprising:
utilizing, by the at least one processor, a machine learning model to ingest a plurality of entity records to produce clusters of the plurality of entity records;
indexing, by the at least one processor, the clusters in a searchable data structure;
determining, by the at least one processor, confidence information for the clusters based at least in part on attributes of at least one data record of the plurality of entity records;
in response to the confidence information, concurrently adjusting, by the at least one processor, at least one parameter of the machine learning model and conditioning a set of data records to be processed;
updating, by the at least one processor, the clusters using the adjusted machine learning model over the conditioned set of data records; and
reindexing, by the at least one processor, the updated clusters.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein mapping clusters comprises employing distinct machine learning models including an entity feature engine configured for coarse feature extraction and a fine feature engine configured for finer resolution, and wherein the fine feature engine is updated instance-dependently in response to the determined category.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1, wherein computing the confidence information comprises applying a method-of-moments estimator to derive at least one confidence bands for at least one cluster based at least in part on a sampled mean and variance of attributes associated with records in the cluster.

### Claim 4 (depends on 1)

The computer-implemented method of claim 1, wherein determining the category comprises identifying an exact match between at least one attribute and an entry of a pre-determined list stored in local memory, and initiating an atomic operation that concurrently adjusts at least one parameter of the machine learning model and conditions the set of data records to be processed.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein indexing the clusters in the searchable data structure comprises parsing cluster data representations using a combination of delimiter-based parsing and fixed-length parsing and storing the parsed representations in an elastic search environment.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, further comprising applying the confidence information as a retrieval constraint to prioritize clusters that satisfy a user-specified tolerance during search and display.

### Claim 7 (depends on 1)

The computer-implemented method of claim 1, wherein adjusting at least one parameter of the machine learning model comprises modifying hidden-layer connection strengths and weights to reduce sensitivity to potential matches when the category indicates higher risk and to increase sensitivity when the category indicates lower risk.

### Claim 8 (depends on 1)

The computer-implemented method of claim 1, further comprising determining a semantic similarity of embeddings of the feature representations and gating inclusion of records into the updated clusters using a similarity threshold in combination with the confidence information.

### Claim 9 (depends on 1)

The computer-implemented method of claim 1, wherein reindexing the updated clusters comprises assigning an index alias to the updated clusters to preserve rollback capability and ensure continuous retrieval during updates.

### Claim 10

A system comprising:
at least one processor;
wherein the at least one processor, upon execution of the programming instructions, is further configured to
utilizing, by the at least one processor, a machine learning model to ingest a plurality of entity records to produce clusters of the plurality of entity records;
indexing, by the at least one processor, the clusters in a searchable data structure;
determining, by the at least one processor, confidence information for the clusters based at least in part on attributes of the plurality of entity records;
in response to the confidence information, concurrently adjusting, by the at least one processor, at least one parameter of the machine learning model and conditioning a set of data records to be processed;
updating, by the at least one processor, the clusters using the adjusted machine learning model over the conditioned set of data records; and
reindexing, by the at least one processor, the updated clusters.

### Claim 11 (depends on 10)

The system of claim 10, wherein mapping clusters comprises employing distinct machine learning models including an entity feature engine configured for coarse feature extraction and a fine feature engine configured for finer resolution, and wherein the fine feature engine is updated instance-dependently in response to the determined category.

### Claim 12 (depends on 10)

The system of claim 10, wherein computing the confidence information comprises applying a method-of-moments estimator to derive at least one confidence band for at least one cluster based at least in part on a sampled mean and variance of attributes associated with records in the cluster.

### Claim 13 (depends on 10)

The system of claim 10, wherein determining the category comprises identifying an exact match between at least one attribute and an entry of a pre-determined list stored in local memory, and initiating an atomic operation that concurrently adjusts at least one parameter of the machine learning model and conditions the set of data records to be processed.

### Claim 14 (depends on 10)

The system of claim 10, wherein indexing the clusters in the searchable data structure comprises parsing cluster data representations using a combination of delimiter-based parsing and fixed-length parsing and storing the parsed representations in an elastic search environment.

### Claim 15 (depends on 10)

The system of claim 10, wherein the at least one processor, upon execution of the programming instructions, is further configured to apply the confidence information as a retrieval constraint to prioritize clusters that satisfy a user-specified tolerance during search and display.

### Claim 16 (depends on 10)

The system of claim 10, wherein adjusting at least one parameter of the machine learning model comprises modifying hidden-layer connection strengths and weights to reduce sensitivity to potential matches when the category indicates higher risk and to increase sensitivity when the category indicates lower risk.

### Claim 17 (depends on 10)

The system of claim 10, wherein the at least one processor, upon execution of the programming instructions, is further configured to determine a semantic similarity of embeddings of the feature representations and gating inclusion of records into the updated clusters using a similarity threshold in combination with the confidence information.

### Claim 18 (depends on 10)

The system of claim 10, wherein reindexing the updated clusters comprises assigning an index alias to the updated clusters to preserve rollback capability and ensure continuous retrieval during updates.

### Claim 19

A computer-implemented method comprising:
determining, by at least one processor, based at least in part on at least one attribute of a plurality of entity records, confidence information for each cluster of a plurality of clusters indexed in a searchable data structure;
wherein the plurality of clusters comprise groupings of the plurality of entity records;
wherein the plurality of data records comprise at least one attribute;
in response to the confidence information, adjusting, by the at least one processor, at least one parameter of the machine learning model;
updating, by the at least one processor, the clusters using the adjusted machine learning model; and
reindexing, by the at least one processor, in the searchable data structure, the updated clusters.

### Claim 20 (depends on 19)

The computer-implemented method of claim 19, wherein computing the confidence information comprises applying a method-of-moments estimator to derive at least one confidence band for at least one cluster based at least in part on a sampled mean and variance of attributes associated with records in the cluster.
