# US 2026/0147750 A1 · Llm-driven approach for structured data extraction from semi-unstructured data with attribute name optimization

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

- Number: US 2026/0147750 A1
- Document ID: US20260147750A1
- Status: Published application
- Kind: A1
- Application number: 18/960,462
- Filed: 2024-11-26
- Published: 2026-05-28
- Earliest claimed: 2024-11-26
- Assignees: Dell Products L.P.
- Applicants: Dell Products L.P.
- Inventors: Leandro Takeshi Hattori; Claudio Romero
- Web page: https://patentgrep.com/patent/US20260147750A1

## CPC

- G06F 16/2372: {Updates performed during offline database operations}
- G06F 16/258: {Data format conversion from or to a database}
- G06F 16/254: {Extract, transform and load [ETL] procedures, e.g. ETL data flows in data warehouses}
- G06F 16/84: Mapping; Conversion
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

One example method includes performing a first data ingestion process that includes (1) receiving a complete SUD (semi-unstructured data) dataset from a data lake and (2) returning a list of prompts for each SUD type with the complete SUD dataset, performing, by an LLM (large language model) a knowledge extraction process on the complete SUD dataset, and the knowledge extraction process uses optimized attribute names and the list of prompts for each SUD to obtain a complete structured dataset, and returning the complete structured dataset to the data lake.

## Claims

### Claim 1

A method for transforming unstructured data to structured data, comprising:
performing a first data ingestion process that comprises (1) receiving a subset of SUD (semi-unstructured data) dataset from a data lake, (2) selecting a first list of prompts for each SUD type based on the subset of the SUD dataset, and (3) returning optimized attribute names based on the first list of prompts and the subset of the SUD dataset;
performing a second data ingestion process that comprises (1) receiving a complete set of the SUD dataset from the data lake and (2) returning a second list of prompts for each SUD type with the complete set of the SUD dataset;
performing, by an LLM (large language model), a knowledge extraction process on the complete set of the SUD dataset, wherein the knowledge extraction process uses the optimized attribute names and the second list of prompts for each SUD to obtain a complete structured dataset for the SUD dataset; and
returning the complete structured dataset to the data lake.

### Claim 2 (depends on 1)

The method as recited in claim 1, wherein the subset of the SUD dataset obtained from the data lake is used to select recommended prompts for knowledge extraction of the complete set of the SUD dataset.

### Claim 3 (depends on 1)

The method as recited in claim 1, wherein the optimized attribute names were obtained by way of a user feedback loop mechanism.

### Claim 4 (depends on 1)

The method as recited in claim 1, wherein the subset of the SUD dataset obtained from the data lake is used to select recommended prompts for knowledge extraction of the complete set of the SUD dataset, while also selecting one or more prompts for each different data type in the subset of the SUD dataset.

### Claim 5 (depends on 1)

The method as recited in claim 1, wherein the subset of the SUD dataset obtained from the data lake is used to select recommended prompts for knowledge extraction of the complete set of the SUD dataset, and the subset of the SUD dataset comprises multiple different data types.

### Claim 6 (depends on 1)

The method as recited in claim 1, wherein the subset of the SUD dataset obtained from the data lake comprises multiple different data types, and the data types were identified using a zero-shot classifier.

### Claim 7 (depends on 1)

The method as recited in claim 1, wherein creation of the complete structured dataset is performed using a non-token-based approach.

### Claim 8 (depends on 1)

The method as recited in claim 1, wherein the LLM comprises an open-source LLM.

### Claim 9 (depends on 1)

The method as recited in claim 1, wherein the optimized attribute names ensure that a view of the complete structured dataset accurately reflects preferences of a user based on whose input the optimized attribute names were generated.

### Claim 10 (depends on 1)

The method as recited in claim 1, wherein the optimized attribute names are generated based on attribute names initially provided by a user.

### Claim 11

A non-transitory storage medium having stored therein instructions that are executable by one or more hardware processors to perform operations comprising:
performing a first data ingestion process that comprises (1) subset of SUD (semi-unstructured data) dataset from a data lake, (2) selecting a first list of prompts for each SUD type based on the subset of the SUD dataset, and (3) returning optimized attribute names based on the first list of prompts and the subset of the SUD dataset;
performing a second data ingestion process that comprises (1) receiving a complete set of the SUD dataset from the data lake and (2) returning a second list of prompts for each SUD type with the complete set of the SUD dataset;
performing, by an LLM (large language model), a knowledge extraction process on the complete set of the SUD dataset, wherein the knowledge extraction process uses optimized attribute names and the second list of prompts for each SUD to obtain a complete structured dataset for the SUD dataset; and
returning the complete structured dataset to the data lake.

### Claim 12 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the subset of the SUD dataset obtained from the data lake is used to select recommended prompts for knowledge extraction of the complete set of the SUD dataset.

### Claim 13 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the optimized attribute names were obtained by way of a user feedback loop mechanism.

### Claim 14 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the subset of the SUD dataset obtained from the data lake is used to select recommended prompts for knowledge extraction of the complete set of the SUD dataset, while also selecting one or more prompts for each different data type in the subset of the SUD-subset dataset.

### Claim 15 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the subset of the SUD dataset obtained from the data lake is used to select recommended prompts for knowledge extraction of the complete set of the SUD dataset, and the subset of the SUD dataset comprises multiple different data types.

### Claim 16 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the subset of the SUD dataset obtained from the data lake comprises multiple different data types, and the data types were identified using a zero-shot classifier.

### Claim 17 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein creation of the complete structured dataset is performed using a non-token-based approach.

### Claim 18 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the LLM comprises an open-source LLM.

### Claim 19 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the optimized attribute names ensure that a view of the complete structured dataset accurately reflects preferences of a user based on whose input the optimized attribute names were generated.

### Claim 20 (depends on 11)

The non-transitory storage medium as recited in claim 11, wherein the optimized attribute names are generated based on attribute names initially provided by a user.
