METHOD AND SYSTEM FOR GENERATING QUERIES TO DATABASE IN NATURAL LANGUAGE Russian patent published in 2025 - IPC G06F16/2455 

FIELD: information technology.

SUBSTANCE: result is achieved through steps of: obtaining user database data, including at least: database structure, description of tables and columns, examples of column values; indexing the obtained user database data into the vector database; converting user database data into a graph; detecting subgraphs in the database containing key-related tables; generating templates of accessing tables by identified subgraphs; performing user database table and field descriptions normalization; machine learning model adapters are trained using a normalized database structure; obtaining a user request containing a database query command in natural language; performing vectorization of the received request and its subsequent processing by the vector database; adding to the context of the user request database structure parameters stored in the vector database; generating a request to a large language model (LLM) based on the processed vector database of the user request.

EFFECT: high accuracy of converting instructions for accessing a database from a natural language to the language of SQL.

5 cl, 4 dwg