Text2SQL System Based On Explicit Relational Encoding And Abstract Syntax Tree Decoding

The rise of big data has stimulated the rapid development of data scale.A large number of high-value data are stored in the database.In the past,when users want to query the database content,they need to write structured query language SQL,and then interact with the database,which brings inconvenience to ordinary users in non professional fields.Moreover,for some complex query conditions,it is easy to make mistakes when writing SQL manually.How to query database freely through natural language has become a new research hotspot.Therefore,this thesis proposes a text2 sql system based on explicit relation coding and abstract syntax tree decoding,which can convert natural language into SQL sequence end-to-end.The main work of this thesis is as follows:(1)Relational Tree SQL model is put forward,which can be effectively complex SQL queries across different areas of table model encoder part introduces the relationship between sensitive attention mechanism,solve the links between natural language and database schema decoder decoding through the predefined SQL syntax tree constraints when the search range,save time cost and improves the accuracy.(2)The text2 sql system is encapsulated and implemented.Due to the limitation of SQL syntax and the powerful semantic generalization ability of the model,the system can be easily migrated to other fields,which can significantly alleviate the lack of data annotation.And through the end-to-end design,the text2 sql system can be quickly integrated into the existing human-computer interaction system.(3)The evaluation and ablation experiments were carried out on the authoritative text2 sql dataset spider and wikisql.By the beginning of October 2020,the public test results of the system on spider have surpassed the solutions of Allen laboratory,National University of Singapore and Korea instant messaging company,and have won the second place in the world,which proves the effectiveness of the system.