Research And Implementation Of Structured Query Generation Method For Natural Language Questions

The rapid development of the Internet in the 21 st century has brought people from the industrial age to the information age.People's daily life will generate a large amount of information,most of which will be collected and stored for later query and use.The development of database technology provides a one-stop service for the storage,management,and retrieval of data information,but using database requires the mastery of professional structured query languages,such as SQL.This kind of human-computer interaction is not friendly enough for most people,and many people do not have the professional skills to use structured query language,making the lack of such skills an obstacle to use databases to retrieve information.With the development of artificial intelligence,human-computer interaction technology has also been fully developed,and natural language interaction as a convenient and friendly way of interaction has attracted more and more research.As the basic technology of natural language interaction,natural language processing can process human natural language in various ways to complete specific tasks.Among them,semantic analysis can convert natural language into a language that can be understood and executed by a computer,which is better unleash the potential of humans to operate machines and reduce the learning costs of humans using machines.This research uses the semantic analysis technology in natural language processing to construct a transformation model based on deep learning methods to transform natural language questions into SQL statements that can be executed by the computer.The model uses the encoder-decoder approach in deep learning as the basic structure and uses a higher abstract dimensional syntax tree as the intermediate representation of natural language and SQL statements.In the encoding stage,the method of pattern connection is used to match and mark the semantic entities mentioned in natural language with the column names in the database,and the deep learning model BERT is used to enhance the representation of the data.In the decoding stage,in order to improve the recognition rate of column names in the database.the dynamic memory block is used to record the column names which have be selected previously.Experimental results show that compared with other models,the method proposed in this study has a certain improvement in the accuracy of generating SQL statements on the Spider dataset,which is close to the performance of the current State-Of-The-Art model.