Method for converting natural language into SQL (Structured Query Language) statement based on deep learning

Abstract

The invention relates to a method for converting a natural language into an SQL (Structured Query Language) statement based on deep learning, in particular to a technology for obtaining a lightweight model by using a knowledge distillation method, and belongs to the technical fields of database intelligent retrieval, question and answer systems and the like. According to the NL2SQL algorithm based on deep learning, coding training is conducted on natural languages of English and Chinese based on a BERT pre-training model to obtain an NL2SQL model, then a knowledge distillation method is used for obtaining a lightweight NL2SQL model, the model is trained according to a provided NL2SQL task data set, and after natural language questions are input, corresponding SQL statements can be inferred; according to the method, Chinese and English NL2SQL tasks can be processed; the calculation resource consumption of the model is reduced, and the reasoning speed is increased.

Description

technical field [0001] The invention relates to a method for converting natural language into SQL statements based on deep learning, in particular to a light-weight model technology using a knowledge distillation method, and belongs to the technical fields of database intelligent retrieval, question answering system and the like. Background technique [0002] In real life, people are usually accompanied by the generation of massive data in the process of work and life, such as file management, online shopping, banking and so on. In this process, a large amount of structured and semi-structured data is generated, which is generally stored, managed and utilized effectively by using relational databases. Structured Query Language SQL (StructuredQuery Language) is a general-purpose, highly functional database manipulation language. Therefore, the process of querying data in a relational database usually requires interaction by using the SQL query language. However, using SQL q...

AI Technical Summary

Problems solved by technology

[0007] (1) The algorithm is only for the English WikiSQL data set, and can only solve the simple SQL statement generation, only select a single column, and only handle a single AND or OR in the conditional clause;

[0008] (2) X-SQL can only effectively handle English NL2SQL tasks, and the accuracy rate of Chinese NL2SQL tasks is low;

[0009] (3) X-SQL uses a large-scale NLP pre-training model MT-DNN as an encoder. The training time is long, the model calculation consumes a lot of resources, the reasoning speed is not fast enough, and the time complexity and space complexity of the algorithm need to be improved.

Images