Granted Patent B1

US 12,566,757 · App. 18/898,734 · Granted Mar 3, 2026

Inventors: Yi Ming Wang (Xian, CN); Rui Han (Xian, CN); Mu Dan Cao (Beijing, CN); Jun Guo (Xian, CN); Sen Liang (Xian, CN); Deng Xin Luo (Xian, CN); Yu Zui You (Ningbo, CN)

Assignee: International Business Machines Corporation

G06F16/24522G06N3/092

Abstract

Claims (111)

1 . A method comprising:

extracting database schema from a target database;

inputting the database schema into a generative model;

training the generative model using a Monte-Carlo method;

training a discriminator model to distinguish generated and real input data by;

fixing parameters of the generative model;

applying a first label to real input data;

applying a second label to generated input data;

merging the labeled real input data and the labeled generated input data into a common dataset;

providing the common dataset as input to the discriminator model yielding a set of results;

determining loss function values according to the set of results and the respective labels of the common dataset; and

updating discriminator model parameters according to the loss function values;

determining a loss value for generated input data;

training the generative model using the loss value model;

receiving, over a network, input text from a user;

automatically generating, using the generative model, an output using the input text; and

providing, over the network, the output to the user.

2 . The computer implemented method according to claim 1 , further comprising:

receiving, over the network, a query;

determining, by the discriminator model, a status of the query; and

providing, over the network, the status to the user.

3 . The computer implemented method according to claim 1 , wherein the output comprises an SQL query, a table schema, and data information.

4 . The computer implemented method according to claim 1 , wherein determining a loss value for generated input data comprises:

receiving incomplete generated input data from a temporary generative model;

generating a plurality of remainder data to yield a plurality of completed generated input data;

evaluating each of the plurality of completed generated input data;

determining a reward value for each of the completed generated input data according to the evaluation;

determining an expectation of reward values according to the reward values for the plurality of completed generated input data as a loss function value; and

updating temporary generative model parameters according to the loss function value.

5 . The computer implemented method according to claim 1 , further comprising:

updating generative model parameters using temporary generative model parameters;

receiving complete input data from the generative model;

evaluating the complete input data using the discriminator model, yielding a loss function value; and

updating the generative model parameters according to the loss function value.

6 . The computer implemented method according to claim 1 , wherein training the generative model comprises:

receiving a training epoch threshold value from a user over the network; and

training the generative model for the training epoch threshold value number of epochs.

7 . A computer program product for generating structured query language queries, the computer program product comprising one or more computer readable storage media and collectively stored program instructions on the one or more computer readable storage media, the stored program instructions which, when executed, cause one or more computer processors to provide a method including:

extracting database schema from a target database;

inputting the database schema into a generative model;

training the generative model using a Monte-Carlo method;

training a discriminator model to distinguish generated and real input data by;

fixing parameters of the generative model;

applying a first label to real input data;

applying a second label to generated input data;

merging the labeled real input data and the labeled generated input data into a common dataset;

providing the common dataset as input to the discriminator model yielding a set of results;

determining loss function values according to the set of results and the respective labels of the common dataset; and

updating discriminator model parameters according to the loss function values;

determining a loss value for generated input data;

training the generative model using the loss value yielding a revised generative model;

receiving, over a network, input text from a user;

automatically generating, using the revised generative model, an output using the input text; and

providing, over the network, the output to the user.

8 . The computer program product according to claim 7 , further comprising:

receiving, over the network, a query;

determining, by the discriminator model, a status of the query; and

providing, over the network, the status to the user.

9 . The computer program product according to claim 7 , wherein the output comprises an SQL query, a table schema, and data information.

10 . The computer program product according to claim 7 , wherein determining a loss value for generated input data comprises:

receiving incomplete generated input data from a temporary generative model;

generating a plurality of remainder data to yield a plurality of completed generated input data;

evaluating each of the plurality of completed generated input data;

determining a reward value for each of the completed generated input data according to the evaluation;

determining an expectation of reward values according to the reward values for the plurality of completed generated input data as a loss function value; and

updating temporary generative model parameters according to the loss function value.

11 . The computer program product according to claim 7 , further comprising:

updating generative model parameters using temporary generative model parameters;

receiving complete input data from the generative model;

evaluating the complete input data using the discriminator model, yielding a loss function value; and

updating the generative model parameters according to the loss function value.

12 . The computer program product according to claim 7 , wherein training the generative model comprises:

receiving a training epoch threshold value from a user over the network; and

training the generative model for the training epoch threshold value number of epochs.

13 . A computer system for generating structured query language queries, the computer system comprising:

one or more computer processors;

one or more computer readable storage media; and

stored program instructions on the one or more computer readable storage media for execution by the one or more computer processors, the stored program instructions which, when executed, cause one or more computer processors to provide a method including:

extracting database schema from a target database;

inputting the database schema into a generative model;

training the generative model using a Monte-Carlo method;

training a discriminator model to distinguish generated and real input data by;

fixing parameters of the generative model;

applying a first label to real input data;

applying a second label to generated input data;

merging the labeled real input data and the labeled generated input data into a common dataset;

providing the common dataset as input to the discriminator model yielding a set of results;

determining loss function values according to the set of results and the respective labels of the common dataset; and

updating discriminator model parameters according to the loss function values;

determining a loss value for generated input data;

training the generative model using the loss value model;

receiving, over a network, input text from a user;

automatically generating, using the generative model, an output using the input text; and

providing, over the network, the output to the user.

14 . The computer system according to claim 13 , further comprising:

receiving, over the network, a query;

determining, by the discriminator model, a status of the query; and

providing, over the network, the status to the user.

15 . The computer system according to claim 13 , wherein the output comprises an SQL query, a table schema, and data information.

16 . The computer system according to claim 13 , wherein determining a loss value for generated input data comprises:

receiving incomplete generated input data from a temporary generative model;

generating a plurality of remainder data to yield a plurality of completed generated input data;

evaluating each of the plurality of completed generated input data;

determining a reward value for each of the completed generated input data according to the evaluation;

determining an expectation of reward values according to the reward values for the plurality of completed generated input data as a loss function value; and

updating temporary generative model parameters according to the loss function value.

17 . The computer system according to claim 13 , further comprising:

updating generative model parameters using temporary generative model parameters;

receiving complete input data from the generative model;

evaluating the complete input data using the discriminator model, yielding a loss function value; and

updating the generative model parameters according to the loss function value.

Assignments (1)

References Cited (24)