Text-to-SQL model optimization method and system

Abstract

The embodiment of the invention provides a text-to-SQL model optimization method. The method comprises the following steps: inputting relation characteristics determined from a heterogeneous graph of a text and a line graph corresponding to the heterogeneous graph into a hidden layer; in the hidden layer, iteratively updating node features and edge features through two relational graph attention networks, and inputting the iteratively updated node features to an output layer; and determining the corresponding SQL in the output layer. The embodiment of the invention further provides a text-to-SQL model optimization system. According to the embodiment of the invention, in the encoding task of the heterogeneous graph, the encoding capability of the heterogeneous graph is further improved by introducing the line graph structure of the heterogeneous graph, enriching the semantic features and modeling the topological structure between the edges, and meanwhile, bidirectional modeling is performed on the nodes and the edges, so that the performance is improved in the face of complex connection conditions.

Description

technical field [0001] The invention relates to the field of intelligent speech, in particular to a method and system for optimizing a text-to-SQL model. Background technique [0002] The text-to-SQL task aims to convert natural language questions into corresponding SQL query statements given the corresponding database schema. [0003] In order to implement text-to-SQL tasks, text-to-SQL models such as SQLNet, TypeSQL, GNNSQL, EditSQL, and RATSQL are usually used. [0004] These models are all designed to jointly encode the input natural language text and database schema, and then convert it into the target output SQL program in an end-to-end manner. The database schema includes the name of each data table in the database, the name of each column in the table, the set of primary keys of the data table, and the foreign key connection information between the data tables. The tables and columns in the database are collectively referred to as schema items, and they can form a ...

AI Technical Summary

Problems solved by technology

[0014] In order to at least solve the problems in the prior art, these methods ignore the connection information between the edges in the directed graph, either only consider the local relationship, or use a fully connected graph to only consider the global relationship, ignoring the local relationship. stressed issue

Images