Contextual graph dynamics for natural language to SQL conversion
A method for training a model to generate a structured query language (SQL) output using a natural language input includes receiving a training data set that includes natural language queries, corresponding SQL statements, and a corresponding database. The method also includes extracting query-specific schema from the corresponding database. Further, the method includes embedding the natural language queries and the query-specific schema to generate initialized embeddings, where each initialized embedding corresponds to a natural language query of the natural language queries, and where the corresponding SQL statement is one of the corresponding SQL statements; generating subgraphs based on the initialized embeddings and the natural language queries; training a graph neural network (GNN) using the subgraphs as inputs; generating, using the GNN, refined outputs based on the subgraphs; and training a transformer model using the refined outputs and the natural language queries to obtain a trained model.
1 . A method for training and using a model to generate a structured query language (SQL) output using a natural language input, the method comprising:
training the model by:
receiving a training data set comprising natural language queries, corresponding SQL statements, and a corresponding database;
extracting query-specific schema from the corresponding database;
embedding the natural language queries and the query-specific schema to generate initialized embeddings, wherein each initialized embedding corresponds to a natural language query of the natural language queries, and wherein the corresponding SQL statement is one of the corresponding SQL statements;
generating subgraphs based on the initialized embeddings and the natural language queries;
generating, using a graph neural network (GNN), refined outputs based on the subgraphs, wherein the GNN is trained using the subgraphs as inputs; and
training a transformer model using the refined outputs and the natural language queries to obtain a trained model;
receiving a user input, wherein the user input comprises a user natural language query;
identifying a target database based on the user input;
generating a subgraph input based on the user input and a database schema associated with the target database;
generating, using the subgraph input and the user input as inputs to the trained model, a SQL query; and
performing, based on the user input and the SQL query, an action from an action set.
2 . The method of claim 1 , wherein the target database is the corresponding database.
3 . The method of claim 1 , wherein the action set comprises at least one of the following: presenting the SQL query to the user, executing the SQL query on the target database and presenting a results list to the user, and executing the SQL query on the target database and manipulating the database based on the user input.
4 . The method of claim 1 , wherein training the model further comprises:
refining, via backpropagation, the GNN and transformer model based on a loss function.
5 . The method of claim 1 , wherein the GNN is further trained utilizing a regularization technique and an over-smoothing technique.
6 . The method of claim 1 , wherein the embedding is based on pre-trained word embeddings, metadata associated with the corresponding database, and query context.
7 . The method of claim 1 , wherein the GNN is further trained by, for each subgraph of the subgraphs, dynamically adjusting a number of layers of the GNN based on a number of nodes for the each subgraph.
8 . A method for training a model to generate a structured query language (SQL) output using a natural language input, the method comprising:
receiving a training data set comprising natural language queries, corresponding SQL statements, and a corresponding database;
extracting query-specific schema from the corresponding database;
embedding the natural language queries and the query-specific schema to generate initialized embeddings, wherein each initialized embedding corresponds to a natural language query of the natural language queries, and wherein the corresponding SQL statement is one of the corresponding SQL statements;
generating subgraphs based on the initialized embeddings and the natural language queries;
training a graph neural network (GNN) using the subgraphs as inputs;
generating, using the GNN, refined outputs based on the subgraphs; and
training a transformer model using the refined outputs and the natural language queries to obtain a trained model.
9 . The method of claim 8 , wherein training the model further comprises:
refining, via backpropagation, the GNN and transformer model based on a loss function.
10 . The method of claim 8 , wherein training the GNN further comprises:
utilizing a regularization technique and an over-smoothing technique.
11 . The method of claim 8 , wherein the embedding is based on pre-trained word embeddings, metadata associated with the corresponding database, and query context.
12 . The method of claim 8 , wherein training the GNN further comprises:
for each subgraph of the subgraphs, dynamically adjusting a number of layers of the GNN based on a number of nodes for the each subgraph.
13 . A method for using a model to generate a structured query language (SQL) output using a natural language input, the method comprising:
receiving a user input, wherein the user input comprises a user natural language query;
identifying a target database based on the user input;
generating a subgraph input based on the user input and a database schema associated with the target database;
generating, using the subgraph input and the user input as inputs to the model, a SQL query; performing, based on the user input and the SQL query, an action from an action set;
training the model by:
receiving a training data set comprising natural language queries, corresponding SQL statements, and a corresponding database;
extracting query-specific schema from the corresponding database;
embedding the natural language queries and the query-specific schema to generate initialized embeddings, wherein each initialized embedding corresponds to a natural language query of the natural language queries, and wherein the corresponding SQL statement is one of the corresponding SQL statements;
generating subgraphs based on the initialized embeddings and the natural language queries;
training a transformer model using the subgraphs and the natural language queries to obtain a trained model;
training a graph neural network (GNN) using the subgraphs as inputs, wherein the GNN is used to obtain the trained model; and
refining, via backpropagation, the GNN and transformer model based on a loss function.
14 . The method of claim 13 , wherein the action set comprises at least one of the following: presenting the SQL query to the user, executing the SQL query on the target database and presenting a results list to the user, executing the SQL query on the target database and manipulating the database based on the user input.
15 . The method of claim 13 , wherein the target database is the corresponding database.
16 . The method of claim 13 , wherein training the GNN further comprises utilizing a regularization technique and an over-smoothing technique.
17 . The method of claim 13 , wherein training the GNN further comprises, for each subgraph, dynamically adjusting a number of layers of the GNN based on a number of nodes for the each subgraph.
18 . The method of claim 13 , wherein the embedding is based on pre-trained word embeddings, metadata associated with the corresponding database, and query context.