Research On Database Natural Language Query Model Based On Deep Learning

In a relational database,a large amount of structured information is stored,which is very valuable,but access to a relational database is not easy.It requires users to master a query language such as SQL,and SQL language learning There is a threshold.Natural language is used as the main information expression method in daily life.Combining it with database query technology enables non-professional users to query the information they need in a more convenient way.It has good application prospects and research on natural language query.It has gradually become a research hotspot.The current research in the field of natural language query technology has the following problems: mostly concentrated in the field of English;lack of the positioning process of the target database table,which is inconsistent with the actual application needs;when multiple condition columns and condition values are involved in the natural language question,There may be a value matching error.In order to solve these problems,an end-to-end natural language query locating and transformation model LT-SQL based on deep learning is proposed.The natural language query process is divided into two steps to solve: 1.Target database table positioning;2.SQL keyword generation.Specifically,in step 1,a target database table location method based on the semantic neighbor vector of the question is proposed to find the nearest neighbor of the natural language question in the training data,and the database table of the neighbor question is used as the target database of the current natural language question table;in step 2,the question is divided into multiple sub-tasks to be solved separately,and finally the results of all sub-tasks are combined to generate SQL statements.In order to solve the problem of mismatching conditional columns and conditional values,a corresponding condition is generated for each column of the database table header,and the specified number of conditions are taken out according to the probability as the WHERE clause of the SQL query statement.Experiments were conducted on two Chinese natural language query datasets.The experiment is divided into two parts: 1.Target database table positioning,comparison models include classic classification models and other schemes based on semantic neighbors;2.SQL keywords generation,comparison models include the classic model and the latest model in the field of natural language query.The experimental results show that the proposed natural language query model performs better than other models on real business data,and the combination of the above two modules can achieve a higher natural language query accuracy and achieve the standard of applying in actual systems.