NLP Training Data Augmentation Through Schema-Synchronized SQL
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for augmenting training data for neural networks translating natural language queries to SQL queries require manually created templates or natural language training data, and do not ensure diversity in database schemas.
Innovation Solution
A method that augments training datasets by modifying database schemas through schema modification operations and SQL query synchronization, generating new training data without additional natural language data, using a neural network-based translator.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If template-based methods are used to generate synthetic training data, then data augmentation can be performed, but manual work of defining multiple templates is required and expressions are limited
Solution Approach 1:
The patent uses SQL-to-text models to automatically generate natural language queries from existing SQL queries and database schemas, copying and transforming structured data into natural language training data without requiring manual template creation. This automates the data generation process while maintaining domain-specific accuracy.
Solution Approach 2:
The patent modifies database schemas by applying various transformations (adding columns, renaming tables, changing data types) to generate diverse training datasets. By changing schema parameters systematically, the method generates varied training examples without manual intervention.
2Adaptability or versatility
If SQL-to-text models are used to generate natural language queries, then various natural language expressions can be generated, but it is still difficult to collect or generate various data for training the model
Solution Approach 1:
The patent performs preliminary database schema modifications and SQL query transformations before training the SQL-to-text model. By pre-processing the database schemas and generating corresponding SQL queries with varied structures, the method prepares diverse training data in advance, enabling the model to learn multiple natural language expression patterns.
Solution Approach 2:
The patent dynamically generates training data by applying various schema modification operations (add column, rename table, change data type) to create evolving database structures. This dynamic approach continuously produces new training examples with different natural language expressions, preventing data exhaustion.
3Quantity of substance
If existing training data augmentation methods are used, then training data quantity can be increased, but diversity of database schemas is not ensured
Solution Approach 1:
The patent systematically changes database schema parameters through defined transformation operations including adding columns, renaming tables, changing data types, and modifying relationships. Each transformation creates a new schema variant, ensuring diverse database structures in the training dataset while maintaining data quantity growth.
Data Source
AI summary
A learning data set enhancing method comprises the steps of: determining a first initial learning data set comprising a first learning-purpose natural language query, database information regarding the database, and a first learning-purpose SQL query corresponding to the first learning-purpose natural language query; generating first novel database information regarding a first novel database having a different schema from the database by applying a first schema modification operation to the database; generating a first novel learning-purpose SQL query regarding the first novel database by applying a first SQL query synchronization operation corresponding to the first schema modification operation to the first learning-purpose SQL query; and determining a first novel learning data set comprising the first learning-purpose natural language query, the first novel database information, and the first novel learning-purpose SQL query.