Systems and methods for synthetic database query generation

Patent metadata

Field	Value
Publication number	US-11513869-B2
Application number	US-201916298463-A
Country	US
Kind code	B2
Filing date	Mar 11, 2019
Priority date	Jul 6, 2018
Publication date	Nov 29, 2022
Grant date	Nov 29, 2022

Abstract

A system for returning synthetic database query results. The system may include a memory unit for storing instructions, and a processor configured to execute the instructions to perform operations comprising: receiving a query input by a user at a user interface; determining, based on natural language processing, a type of the query input; determining, based on the received query input and a database language interpreter, an output data format; returning, based on a generation model and the output data format, a result of the query input; providing, to a plurality of training models and based on the determined query type, the query input and the result; and training the training models, based on the query input and the result.

First claim

What is claimed is:

1. A system for training models for outputting synthetic database query results, the system comprising:
at least one memory unit for storing instructions; and
at least one processor configured to execute the instructions to perform operations comprising:
receiving a first query input entered by a user at a user interface; determining a type of the first query input;
generating a synthetic dataset using a dataset generator comprising a trained generative adversarial network, the synthetic dataset:
differing by at least a predetermined amount from a reference dataset according to a similarity metric; and
comprising synthetic data portions generated by:
determining a class of sensitive data portions in the first query input;
selecting a subclass of sensitive data portions within the class based on a distribution model;
generating synthetic data portions using a subclass-specific model trained to generate synthetic values for the selected subclass and not for other subclasses within the class; and
replacing the sensitive data portions with the synthetic data portions;

based on the determined first query input type, providing the first query input and the synthetic dataset to a plurality of training models;
training the plurality of training models based on the first query input and the synthetic dataset;
receiving a second query input;
determining a type of the second query input; and
routing the second query input to a selected training model of the plurality of training models based on the determined second query input type and an output format of the selected training model.

2. The system of claim 1 , wherein the operations further comprise generating an expected database result from the first query input.

3. The system of claim 1 , wherein the operations further comprise evaluating, by a model optimizer, performance criteria of the plurality of training models.

4. The system of claim 3 , wherein the performance criteria includes the similarity metric.

5. The system of claim 1 , wherein the operations further comprise:
determining, based on the first query input and a database language interpreter, an output data format;
returning, based on a generation model and the output data format, a synthetic result of the first query input;
extracting information from the generation model; and
displaying the extracted information on the user interface.

6. The system of claim 1 , wherein the reference dataset includes data represented in least one of a structured data format, a semi-structured data format, or an unstructured data format.

7. The system of claim 1 , wherein the first query input comprises customer financial information.

8. The system of claim 7 , wherein determining the type of the second query input includes using a database language interpreter to interpret a language associated with the second query input.

9. The system of claim 7 , wherein the operations further comprise predicting the subclass of sensitive data portions.

10. The system of claim 9 , wherein the subclass of sensitive data portions is predicted based on at least one additional subclass.

11. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform operations, the operations comprising:
receiving a first query input entered by a user at a user interface;
determining a type of the first query input;
generating a synthetic dataset using a dataset generator comprising a trained generative adversarial network, the synthetic dataset:
differing by at least a predetermined amount from a reference dataset according to a similarity metric; and
comprising synthetic data portions generated by:
determining a class of sensitive data portions in the first query input;
selecting a subclass of sensitive data portions within the class based on a distribution model:
generating synthetic data portions to replace the sensitive data portions using a subclass-specific model trained to generate synthetic values for the selected subclass and not for other subclasses within the class;

based on the determined first query input type, providing the first query input and the synthetic dataset to a plurality of training models;
training the plurality of training models based on the first query input and the synthetic dataset;
receiving a second query input;
determining a type of the second query input; and
routing the second query input to a selected training model of the plurality of training models based on the determined second query input type and an output format of the selected training model.

12. A computer-implemented method for training models for outputting synthetic database query results, the method comprising:
receiving a first query input entered by a user at a user interface;
determining a type of the first query input;
generating a synthetic dataset using a dataset generator comprising a trained generative adversarial network, the synthetic dataset:
differing by at least a predetermined amount from a reference dataset according to a similarity metric; and
comprising synthetic data portions generated by:
determining a class of sensitive data portions in the first query input; and
selecting a subclass of sensitive data portions within the class based on a distribution model;
generating synthetic data portions to replace the sensitive data portions using a subclass-specific model trained to generate synthetic values for the selected subclass and not for other subclasses within the class;

based on the determined first query input type, providing the first query input and the synthetic dataset to a plurality of training models;
training the plurality of training models based on the first query input and the synthetic dataset;
receiving a second query input;
determining a type of the second query input; and
routing the second query input to a selected training model of the plurality of training models based on the determined second query input type and an output format of the selected training model.

13. The computer-implemented method of claim 12 , wherein the method further comprises generating an expected database result from the first query input.

14. The computer-implemented method of claim 12 , wherein the method further comprises evaluating, by a model optimizer, performance criteria of the plurality of training models.

15. The computer-implemented method of claim 14 , wherein the performance criteria includes the similarity metric.

16. The computer-implemented method of claim 12 , wherein the method further comprises:
determining, based on the first query input and a database language interpreter, an output data format;
returning, based on a generation model and the output data format, a synthetic result of the first query input;
extracting information from the generation model; and
displaying the extracted information on the user interface.

17. The computer-implemented method of claim 12 , wherein the reference dataset includes data represented in at least one of a structured data format or a semi-structured data format.

18. The computer-implemented method of claim 2 , herein the first query input comprises customer financial information.

19. The computer-implemented method of claim 12 , wherein determining the type of the second query input includes using a database language interpreter to interpret a language associated with the second query input.

20. The comput

Assignees

• Capital One Services Llc

Inventors

• Goodsitt Jeremy
• Walters Austin
• Pham Vincent
• Abdi Taghi Abad Fardin

Classifications

• G06F9/541 Primary

  via adapters, e.g. between incompatible applications · CPC title

• G06F21/6254

  by anonymising data, e.g. decorrelating personal data from the owner's identification · CPC title

• G06T11/10

  Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77) · CPC title

• G06F18/2193

  based on specific statistical tests · CPC title

• G06F18/285

  Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system · CPC title