Generating hierarchical queries from natural language

Patent metadata

Field	Value
Publication number	US-12475112-B2
Application number	US-202318486603-A
Country	US
Kind code	B2
Filing date	Oct 13, 2023
Priority date	Oct 13, 2023
Publication date	Nov 18, 2025
Grant date	Nov 18, 2025

Abstract

The present disclosure relates to generating hierarchical queries from text queries. A query generation system is configured to encode a text query to obtain a text embedding. Then, the system may select a field of a data schema by comparing the text embedding to a field embedding corresponding to the field. Subsequently, the system may generate a hierarchical query including a value corresponding to the selected field. Some implementations of the system may further include one or more formatting models configured to format values included in the text query.

First claim

What is claimed is:

1 . A method implemented by a computing device including at least one processor, the method comprising:
encoding, using a text encoder, a text query to obtain a text embedding, wherein the text query comprises a natural language text query; selecting, using a key phrase mapping component, a field of a data schema by comparing the text embedding to a field embedding corresponding to the field, wherein the data schema comprises one or more hierarchical rules constraining the use of the field for querying data based on a hierarchy of attributes; and generating, using a query composer, a hierarchical query including a value corresponding to the selected field by predicting a next word in a sequence of words based on a positional encoding for the next word, wherein the hierarchical query comprises a hierarchy of nested descriptors corresponding to the hierarchy of attributes in a database.

2 . The method of claim 1 , further comprising:
extracting, using a machine learning model, a key phrase from the text query, wherein the text embedding represents the key phrase.

3 . The method of claim 2 , further comprising:
generating, using the machine learning model, a hierarchy of key phrases for the text query, wherein the key phrase corresponds to a node in the hierarchy.

4 . The method of claim 2 , further comprising:
generating, using the machine learning model, a list of key phrases for the text query, wherein the key phrase corresponds to an element of the list.

5 . The method of claim 1 , further comprising:
generating a value for the field using an operator detection model, wherein the value includes a mathematical operator.

6 . The method of claim 5 , further comprising:
applying an operator filter to the text query, wherein the operator detection model is used based on the operator filter.

7 . The method of claim 1 , further comprising:
generating a value for the field using a value detection model, wherein the value represents a discrete value in the text query.

8 . The method of claim 1 , further comprising:
generating a value for the field using a Boolean detection model, wherein the value comprises a Boolean value.

9 . The method of claim 1 , further comprising:
generating a value for the field using a date generation model, wherein the value comprises a date.

10 . The method of claim 1 , further comprising:
generating a value for the field using a number formatting model, wherein the value comprises a number.

11 . The method of claim 1 , further comprising:
generating a segment of user data based on the hierarchical query.

12 . A method implemented by a computing device including at least one processor, the method comprising:
initializing a machine learning model; obtaining training data including a text query and a ground-truth hierarchy of key phrases corresponding to the text query, wherein the text query comprises a natural language text query; and training the machine learning model to generate, by predicting a next word in a sequence of words based on a positional encoding for the next word, a hierarchy of key phrases based on the text query using the training data.

13 . The method of claim 12 , wherein:
the text query comprises a search query, and the ground-truth hierarchy of key phrases corresponds to fields in a hierarchical schema.

14 . The method of claim 12 , further comprising:
sorting the training data based on length or hierarchical depth, wherein the training is based on the sorted training data.

15 . An apparatus comprising:
at least one processor; at least one memory including instructions executable by the at least one processor; the apparatus further comprising a machine learning model including parameters stored on the at least one memory and configured to generate a hierarchy of key phrases from a text query by predicting a next word in a sequence of words based on a positional encoding for the next word, wherein the text query comprises a natural language text query; a key phrase mapping component configured to identify a field of a data schema, wherein the field corresponds to a key phrase of the hierarchy of key phrases, and wherein the data schema comprises one or more hierarchical rules constraining the use of the field for querying databased on a hierarchy of attributes; and a query composer configured to generate a hierarchical query including a value corresponding to the field, wherein the hierarchical query comprises a hierarchy of nested descriptors corresponding to the hierarchy of attributes in a database.

16 . The apparatus of claim 15 , further comprising:
a text encoder configured to generate a text embedding from the text query.

17 . The apparatus of claim 15 , further comprising:
an operator detection model configured to identify a mathematical operator in the text query.

18 . The apparatus of claim 15 , further comprising:
a value detection model configured to identify a discrete value in the text query.

19 . The apparatus of claim 15 , further comprising:
a Boolean detection model configured to identify a negation in the text query.

20 . The apparatus of claim 15 , further comprising:
a date generation model configured to identify a date from the text query.

Assignees

• Adobe Inc

Inventors

• Chan Yeuk-Yin
• Bursztyn Victor S
• Koh Eunyee
• Ross Nathan
• Holtcamp Vasanthi

Classifications

• G06F16/2423

  Interactive query statement specification based on a database schema · CPC title

• G06F16/243 Primary

  Natural language query formulation · CPC title