NL2SQ analysis method based on deep learning

Abstract

The invention belongs to the technical field of intelligent search, in particular to an NL2SQ analysis method based on deep learning, and provides the following scheme that an NL2SQL analysis model is included, the NL2SQL analysis model comprises an encoder, the encoder is an NLP pre-training model, the NL2SQL analysis model constructs a network architecture adaptive to an NL2SQL task, and the NL2SQ analysis model is used for analyzing the NL2SQ task. And the NL2SQL analysis model unifies the annotation formats of the public data and the self-annotation data. According to the method, [SEL], [GRP], [ORD] and special marks are added through model input to assist a model analysis method, a two-time calculation architecture of the model and an information fusion mode between two-time prediction, the set network architecture is not limited to a source data form, when data changes, the network structure does not need to be changed, the network architecture basically comprises all basic grammar of SQL, coverage is wider, and the method is suitable for large-scale popularization and application. The landing requirements are better met.

Description

technical field [0001] The present invention relates to the technical field of intelligent search, in particular to an NL2SQ analysis method based on deep learning. Background technique [0002] In intelligent search, the process of computers understanding user query intentions has become a research hotspot in the industry. Before understanding user intentions, it is first necessary to convert natural language into an executable program language that computers can understand and generate accurate statement semantics. [0003] NaturalLanguagetoSQL (NL2SQL) is a method to convert the user's natural language statement into a computer-readable, runnable, and semantic representation that conforms to computer rules. NL2SQL aims to convert natural language into database structured query language (NL2SQL), and natural The key to converting the language into SQL is to parse out what the corresponding parts of the above structured statements are according to the natural language (Quer...

AI Technical Summary

Problems solved by technology

[0009] The current mainstream method based on the background technology is to achieve the purpose of analysis through deep learning, but the current deep learning network framework has the following problems: some parts of the network architecture are heavily dependent on the distribution of data, when the data source changes , the model is not applicable, and the network architecture needs to be readjusted for training; the use of the network architecture for the table column name is only part of the input, and the model is allowed to learn the correlation between the query and the column name, and there is no more consideration for the table column name Meaning; the currently public network architectures are all experiments on a single table, and the tasks do not involve the GROUP, HAVING, and multi-table joint query (JOIN) in the above standard structure, which are difficult but commonly used in business. Grammar, so Due to the technical problem of being unable to evaluate the effect of the current public network, the present invention proposes a deep learning-based NL2SQ analysis method

Images