Granted Patent B1

US 12,481,652 · App. 18/932,150 · Granted Nov 25, 2025

Inventor: Chenghung James Pan (Palo Alto, CA)

Assignee: Palo Alto Networks, Inc.

G06F16/2452G06F16/243G06F40/186G06F40/284G06F40/40

Abstract

Claims (58)

1 . A method comprising:

based on receiving a first natural language query for data in one or more databases related to a domain, identifying one or more identifiers of one or more entities in the first natural language query;

determining a mapping between the one or more identifiers in the first natural language query and one or more entity attributes according to a mapping table between entity identifiers and entity attributes for the domain;

replacing, in the first natural language query, the one or more identifiers with mapped ones of the one or more entity attributes indicated in the mapping to obtain a second natural language query;

prompting a foundation model with a first prompt to obtain first one or more database queries for the one or more databases, wherein the first prompt comprises the second natural language query and task instructions to generate first one or more database queries for the one or more databases based on the second natural language query;

replacing those of the one or more entity attributes in the first one or more database queries with corresponding ones of the one or more identifiers according to the mapping to generate second one or more database queries; and

querying the one or more databases with the second one or more database queries to obtain results based on which to generate a response to the first natural language query.

2 . The method of claim 1 further comprising:

retrieving one or more examples from a knowledge base according to similarity with the second natural language query; and

populating a prompt template with the one or more examples and the second natural language query to generate the first prompt.

3 . The method of claim 2 , wherein the one or more examples comprise example attribute-based natural language queries and corresponding example attribute-based database queries.

4 . The method of claim 2 , wherein retrieving the one or more examples from the knowledge base comprises:

generating an embedding of the second natural language query; and

querying the knowledge base with the embedding.

5 . The method of claim 2 , further comprising populating the knowledge base with example attribute-based natural language queries and corresponding example attribute-based database queries, wherein populating the knowledge base comprises:

obtaining example identifier-based natural language queries and corresponding example identifier-based database queries; and

replacing identifiers of entities with corresponding entity attributes in the example identifier-based natural language queries and the example identifier-based database queries according to the mapping table to obtain the example attribute-based natural language queries and the example attribute-based database queries.

6 . The method of claim 2 , wherein the prompt template comprises a description of one or more schemas for one or more query languages of the one or more databases and indications of which of the one or more query languages correspond to the one or more databases.

7 . The method of claim 1 , wherein the domain comprises Internet of Things devices, wherein the entity attributes comprise at least one of device categories, device vendors, and device models.

8 . The method of claim 6 , wherein the one or more query languages comprise the Structured Query Language.

9 . The method of claim 1 , wherein determining the mapping between the one or more identifiers in the first natural language query and one or more entity attributes comprises,

performing a longest prefix search in the first natural language query for at least one of identifiers and identifier prefixes stored in the mapping table to identify the one or more identifiers; and

mapping the one or more identifiers to corresponding ones of the one or more entity attributes according to the mapping table to generate the mapping.

10 . A non-transitory machine-readable medium having program code stored thereon, the program code comprising instructions to:

based on receiving a first natural language query for data in a database, extract one or more identifiers of one or more entities from the first natural language query;

determine a mapping between the one or more identifiers in the first natural language query and one or more entity attributes according to a mapping table between entity identifiers and entity attributes, wherein the mapping table is based on context of a domain related to the database;

replace, in the first natural language query, the one or more identifiers with mapped ones of the one or more entity attributes indicated in the mapping to obtain a second natural language query;

prompt a foundation model with a first prompt to obtain a first database query for the database, wherein the first prompt comprises the second natural language query and task instructions to generate first database query for the database based on the second natural language query;

replace those of the one or more entity attributes in the first database query with corresponding ones of the one or more identifiers according to the mapping to generate a second database query; and

query the database with the second database query to obtain results based on which to generate a response to the first natural language query.

11 . The machine-readable media of claim 10 , wherein the program code further comprises instructions to:

retrieve one or more examples from a knowledge base according to similarity with the second natural language query; and

populate a prompt template with the one or more examples and the second natural language query to generate the first prompt.

12 . The machine-readable media of claim 11 , wherein the one or more examples comprise example attribute-based natural language queries and corresponding example attribute-based database queries.

13 . The machine-readable media of claim 11 , wherein the instructions to retrieve the one or more examples from the knowledge base comprise instructions to:

generate an embedding of the second natural language query; and

query the knowledge base with the embedding.

14 . The machine-readable media of claim 12 , wherein the program code further comprises instructions to populate the knowledge base with example attribute-based natural language queries and corresponding example attribute-based database queries, wherein the instructions to populate the knowledge base comprise instructions to:

obtain example identifier-based natural language queries and corresponding example identifier-based database queries; and

replace identifiers of entities with corresponding entity attributes in the example identifier-based natural language queries and the example identifier-based database queries according to the mapping table to obtain the example attribute-based database queries and the example attribute-based database queries.

15 . The machine-readable media of claim 12 , wherein the prompt template comprises a description of one or more schemas for a query language of the database.

16 . An apparatus comprising:

a processor; and

a machine-readable medium having instructions stored thereon that are executable by the processor to cause the apparatus to,

scan a first natural language query to identify and replace one or more identifiers of one or more entities with corresponding one or more attributes of the one or more entities in the first natural language query to generate a second natural language query, wherein the first natural language query is for data in a database related to a domain, wherein the instructions to identify and replace the one or more identifiers with the corresponding one or more attributes comprise instructions executable by the processor to cause the apparatus to map the one or more identifiers to the one or more attributes according to a mapping in a mapping table between entity identifiers and entity attributes for the domain;

prompt a foundation model with a first prompt to obtain a first database query for the database, wherein the first prompt comprises the second natural language query and task instructions to generate a first database query for the database based on the second natural language query;

replace those of the one or more attributes in the first database query with corresponding ones of the one or more identifiers according to the mapping to generate a second database query; and

query the database with the second database query to obtain results based on which to generate a response to the first natural language query.

17 . The apparatus of claim 16 , wherein the machine-readable medium further has stored thereon instructions executable by the processor to cause the apparatus to:

retrieve one or more examples from a knowledge base according to similarity with the second natural language query; and

populate a prompt template with the one or more examples and the second natural language query to generate the first prompt.

18 . The apparatus of claim 17 , wherein the one or more examples comprise example attribute-based natural language queries and corresponding example attribute-based database queries.

19 . The apparatus of claim 17 , wherein the instructions to retrieve the one or more examples from the knowledge base comprise instructions executable by the processor to cause the apparatus to:

generate an embedding of the second natural language query; and

query the knowledge base with the embedding.

20 . The apparatus of claim 17 , wherein the machine-readable medium further has stored thereon instructions executable by the processor to cause the apparatus to populate the knowledge base with example attribute-based natural language queries and corresponding example attribute-based database queries, wherein the instructions to populate the knowledge base comprise instructions executable by the processor to cause the apparatus to:

obtain example identifier-based natural language queries and corresponding example identifier-based database queries; and

replace identifiers of entities with corresponding entity attributes in the example identifier-based natural language queries and the example identifier-based database queries according to the mapping table to obtain the example attribute-based natural language queries and the example attribute-based database queries.

Assignments (1)

References Cited (35)