Natural language database interface

Patent metadata

Field	Value
Publication number	US-2024346021-A1
Application number	US-202318301739-A
Country	US
Kind code	A1
Filing date	Apr 17, 2023
Priority date	Apr 17, 2023
Publication date	Oct 17, 2024
Grant date	—

Abstract

The present disclosure provides an approach for training a machine learning model. Embodiments include receiving text comprising a natural language request. Embodiments include providing one or more inputs to a source machine learning model based on the text, wherein the source machine learning model has been trained using source training data corresponding to a plurality of databases. Embodiments include receiving, from the source machine learning model in response to the one or more inputs, a database query in a syntax corresponding to a target database. Embodiments include generating training data for a target machine learning model based on the text and the database query received from the source machine learning model, wherein the target machine learning model has been trained using a smaller amount of training data than the source training data that was used to train the source machine learning model.

First claim

We claim:

1 . A method of training a machine learning model, the method comprising:
receiving text comprising a natural language request; providing one or more inputs to a source machine learning model based on the text, wherein the source machine learning model has been trained using source training data corresponding to a plurality of databases; receiving, from the source machine learning model in response to the one or more inputs, a database query in a syntax corresponding to a target database; generating training data for a target machine learning model based on the text and the database query received from the source machine learning model, wherein the target machine learning model has been trained using a smaller amount of training data than the source training data that was used to train the source machine learning model; and training the target machine learning model based on the training data.

2 . The method of claim 1 , wherein the generating of the training data is further based on user feedback with respect to the database query received from the source machine learning model.

3 . The method of claim 1 , wherein the target machine learning model is used to determine a new query in the syntax corresponding to the target database based on new text comprising a new natural language request.

4 . The method of claim 1 , wherein the source machine learning model was trained by a third party, and wherein internal logic of the source machine learning model is not available for analysis with respect to the database query received from the source machine learning model.

5 . The method of claim 4 , wherein respective internal logic of the target machine learning model is available for analysis.

6 . The method of claim 1 , wherein the source machine learning model utilizes a larger amount of physical computing resources than the target machine learning model.

7 . The method of claim 6 , further comprising determining whether to use the source machine learning model or the target machine learning model for subsequently-received text based on a comparison of the subsequently-received text with the text.

8 . The method of claim 7 , wherein the comparison is based on generating embeddings of the subsequently-received text and the text.

9 . The method of claim 1 , further comprising determining whether to stop using the source machine learning model based on a determined accuracy of the target machine learning model.

10 . A system for training a machine learning model, comprising:
at least one memory; and at least one processor coupled to the at least one memory, the at least one processor and the at least one memory configured to:
receive text comprising a natural language request;
provide one or more inputs to a source machine learning model based on the text, wherein the source machine learning model has been trained using source training data corresponding to a plurality of databases;
receive, from the source machine learning model in response to the one or more inputs, a database query in a syntax corresponding to a target database;
generate training data for a target machine learning model based on the text and the database query received from the source machine learning model, wherein the target machine learning model has been trained using a smaller amount of training data than the source training data that was used to train the source machine learning model; and
train the target machine learning model based on the training data.

11 . The system of claim 10 , wherein the generating of the training data is further based on user feedback with respect to the database query received from the source machine learning model.

12 . The system of claim 10 , wherein the target machine learning model is used to determine a new query in the syntax corresponding to the target database based on new text comprising a new natural language request.

13 . The system of claim 10 , wherein the source machine learning model was trained by a third party, and wherein internal logic of the source machine learning model is not available for analysis with respect to the database query received from the source machine learning model.

14 . The system of claim 13 , wherein respective internal logic of the target machine learning model is available for analysis.

15 . The system of claim 10 , wherein the source machine learning model utilizes a larger amount of physical computing resources than the target machine learning model.

16 . The system of claim 15 , wherein the at least one processor and the at least one memory are further configured to determine whether to use the source machine learning model or the target machine learning model for subsequently-received text based on a comparison of the subsequently-received text with the text.

17 . The system of claim 16 , wherein the comparison is based on generating embeddings of the subsequently-received text and the text.

18 . The system of claim 10 , wherein the at least one processor and the at least one memory are further configured to determine whether to stop using the source machine learning model based on a determined accuracy of the target machine learning model.

19 . A non-transitory computer readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to:
receive text comprising a natural language request; provide one or more inputs to a source machine learning model based on the text, wherein the source machine learning model has been trained using source training data corresponding to a plurality of databases; receive, from the source machine learning model in response to the one or more inputs, a database query in a syntax corresponding to a target database; generate training data for a target machine learning model based on the text and the database query received from the source machine learning model, wherein the target machine learning model has been trained using a smaller amount of training data than the source training data that was used to train the source machine learning model; and train the target machine learning model based on the training data.

20 . The non-transitory computer readable medium of claim 19 , wherein the generating of the training data is further based on user feedback with respect to the database query received from the source machine learning model.

Assignees

• Vmware Inc

Inventors

• Xuan Chaoting
• Chawla Ravish
• Stuntebeck Erich

Classifications

• G06F16/24522 Primary

  Translation of natural language queries to structured queries · CPC title

• G06F40/20 Primary

  Natural language analysis (semantic analysis of natural language G06F40/30) · CPC title