# US 2026/0228270 A1 · Using large language models based on different searches

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

- Number: US 2026/0228270 A1
- Document ID: US20260228270A1
- Status: Published application
- Kind: A1
- Application number: 19/046,526
- Filed: 2025-02-06
- Published: 2026-08-06
- Earliest claimed: 2025-02-06
- Assignees: Honeywell International Inc.
- Applicants: Honeywell International Inc.
- Inventors: Nagarjuna Reddy Chevuru; Mamata Vishvanath Lohar; Manish Manohar Karlekar
- Web page: https://patentgrep.com/patent/US20260228270A1

## CPC

- G06F 16/3344: {using natural language analysis}
- G06F 16/383: using metadata automatically derived from the content
- G06F 16/313: {Selection or weighting of terms for indexing}
- G06F 16/3329: Natural language query formulation
- G06F 16/3334: {Selection or weighting of terms from queries, including natural language queries}
- G06F 16/9024: {Graphs; Linked lists (G06F16/9027 takes precedence)}
- G06F 40/216: using statistical methods
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G06F 40/30: Semantic analysis
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Approaches for generating a response for a user based on different searches on a set of ranked documents by using a large-language model. In an example, a query received from a user is processed to generate a query embedding, a term-based query, and a semantic query. Once generated, each of the query embedding, the term-based query and the semantic query may be executed on to determine a first set, a second set and a third set of documents from a document collection. The documents thus obtained are scored to identify set of ranked documents. Thereafter, the ranked documents may be used by a large-language model to generate a response for the user based on the set of ranked documents.

## Claims

### Claim 1

A system comprising:
a processor; and
a machine-readable storage medium comprising instructions executable by the processor to:
receive a query from a user;
process the query received from the user, wherein to process the query, the processor is to generate:
a query embedding, wherein the query embedding is an encoded representation corresponding to keywords present in the query;
a term-based query, wherein the term-based query is to execute on a repository comprising an inverted index; and
a semantic query, wherein the semantic query is to query a data structure representing information stored in a knowledge graph comprising a plurality of nodes linked by edges, wherein the edges represent an association between each corresponding node;
execute each of the query embedding, the term-based query and the semantic query to determine a first set, a second set and a third set of documents from a document collection;
score the set of documents to identify set of ranked documents; and
using a large-language model, generate a response for the user based on the set of ranked documents.

### Claim 2 (depends on 1)

The system as claimed in claim 1, wherein to score the set of documents, the instructions are to:
determine a score for each of the documents in the first set, the second set, and the third set;
determine a final rank based on score of each of the documents in the first set, the second set, and the third set, wherein the final rank is determined based on a weight factor attributed to the score of each of the documents in the first set, the second set, and the third set; and
based on a predefined condition and the final rank determined for each document, identify the set of ranked documents.

### Claim 3 (depends on 2)

The system as claimed in claim 2, wherein the weight factor attributed to the score is determined based on type of the query received from the user.

### Claim 4 (depends on 1)

The system as claimed in claim 1, wherein the instructions when executed are to further cause executing the query embedding on an embedding repository, wherein the embedding repository comprises embeddings depicting semantic context and relationships between terms present in one or more documents in the document collection.

### Claim 5 (depends on 4)

The system as claimed in claim 4, wherein the query embedding in the embedding repository, the inverted index, and knowledge graph, are populated based on the document collection.

### Claim 6 (depends on 5)

The system as claimed in claim 5, wherein to populate the embedding repository, the inverted index, and knowledge graph, the instructions when executed are to:
process each of the documents within the document collection to obtain segmented document portions;
convert content of each of the segmented document portions into text-based content;
process the text-based content to identify a set of keywords from the text-based content;
utilize the set of keywords to form a chunk; and
generate, based on the chunk, one of the embeddings in the embedding repository, the inverted index, and the knowledge graph.

### Claim 7 (depends on 1)

The system as claimed in claim 1, wherein to generate the knowledge graph, the instruction are to:
process the chunks to generate a plurality of tokens;
identify terms of interest, concepts, and relationships between various terms of interest; and
generate the knowledge graph in a machine readable format.

### Claim 8 (depends on 1)

The system as claimed in claim 1, wherein the inverted index comprises a mapping between a select term present within one or more constituent documents within the collection of document, and the corresponding one or more constituent documents in which the select term is present.

### Claim 9

A method comprising:
obtaining a first set of documents from a document collection by executing a query embedding in an embedding repository, wherein the query embedding is an encoded representation corresponding to keywords present in a query received from a user;
obtaining a second set of documents from the document collection by executing a term-based query on a repository comprising an inverted index, wherein the term-based query is based on the query received from the user;
obtaining a third set of documents from the document collection by executing a semantic query for querying a data structure representing information stored in a knowledge graph comprising a plurality of nodes linked by edges, and wherein the knowledge graph is based on the document collection;
scoring each of the documents in the first set, the second set, and the third set;
determining a final rank based on scoring of the documents in the first set, the second set, and the third set, wherein the final rank is determined based on a weight factor attributed to the score of each of the documents in the first set, the second set, and the third set; and
generating a response for the user based on the set of ranked documents in response to the query, using a large-language model.

### Claim 10 (depends on 9)

The method as claimed in claim 9, wherein obtaining the first set of documents comprises:
comparing the query embedding with an embedding within an embedding repository, wherein the embeddings encode semantic context and relationships between terms present in the document collection;
determining a semantic similarity score for each of the documents in the document collection, based on the determining; and
compiling a set of documents from the document collection as the first set based on the semantic similarity score and a predefined semantic condition.

### Claim 11 (depends on 9)

The method as claimed in claim 9, wherein obtaining the second set of documents comprises:
comparing a parameter specified in the term-based query with a term present in the inverted index to identify a set of documents as the second set.

### Claim 12 (depends on 9)

The method as claimed in claim 9, wherein obtaining the third set of documents comprises:
comparing a query object specified within the semantic query with document object specified with corresponding nodes of the knowledge graph, wherein the document object are determined based on one or more documents within the document collection;
for nodes matching the query object, determining a priority score based on direct connection between interrelated nodes, path lengths between nodes, number of times the document object appears within the knowledge graph, or combination thereof; and
compiling another set of documents from the document collection as the second set based on the priority score, a predefined semantic condition, or combination thereof.

### Claim 13 (depends on 9)

The method as claimed in claim 9, wherein the weight factor attributed to the score is determined based on type of the query received from the user, the type of the query being one of full-text search, semantic search, and knowledge graph search.

### Claim 14 (depends on 13)

The method as claimed in claim 13, wherein the type of the query is identified based on executable instructions implementing a predefined criteria.

### Claim 15 (depends on 9)

The method as claimed in claim 9, wherein the query embedding in the embedding repository, the inverted index, and knowledge graph, are populated based on the document collection, by:
processing each of the documents within the document collection to obtain segmented document portions;
converting content of each of the segmented document portions into text-based content;
parsing the text-based content to identify a set of keywords from amongst the text-based content;
generating chunks based on the set of keywords; and
generate, based on the chunk, one of the embeddings in the embedding repository, the inverted index, and the knowledge graph.

### Claim 16

A non-transitory computer-readable medium comprising instructions, the instructions being executable by a processing resource to:
in response to processing a query received from the user, generate:
a query embedding, wherein the query embedding is an encoded representation corresponding to keywords present in the query;
a term-based query caused to execute on a repository comprising an inverted index; and
a semantic query caused to query a data structure representing information stored in a knowledge graph comprising a plurality of nodes linked by edges, wherein the edges represent an association between each corresponding nodes;
further cause to execute each of the query embedding, the term-based query and the semantic query to determine a first set, a second set and a third set of documents from a document collection;
generate a list indicating a set of ranked documents; and
using a large-language model, generate a response for the user based on the set of ranked documents.

### Claim 17 (depends on 16)

The non-transitory computer-readable medium of claim 16, wherein the instructions are further executable to:
determine a score for each of the documents in the first set, the second set, and the third set;
arrange each of the documents in the first set, the second set, and the third set based on the score;
determine a final rank based on score of each of the documents in the first set, the second set, and the third set, wherein the final rank is determined based on a weight factor attributed to the score of each of the documents in the first set, the second set, and the third set; and
based on a predefined condition and the final rank determined for each document, identify the set of ranked documents.

### Claim 18 (depends on 17)

The non-transitory computer-readable medium of claim 17, wherein the weight factor attributed to the score is determined based on type of the query received from the user.

### Claim 19 (depends on 16)

The non-transitory computer-readable medium of claim 16, wherein the instructions are further executable to cause execution of the query embedding on an embedding repository, wherein the embedding repository comprises embeddings depicting semantic context and relationships between terms present in the document collection.

### Claim 20 (depends on 16)

The non-transitory computer-readable medium of claim 16, wherein the query embedding in the embedding repository, the inverted index, and knowledge graph, are populated based on the document collection.
