Optimizing natural language interface for relational database

Abstract

This paper is research on developing a natural language interface for the relational databases, (RDBMS), which takes in the natural language in the form of english like statements as inputs from the operator or user and generates answers in the form of SQL(Structured Query Language) queries. Firstly, the input in the form of natural language is parsed using a semantic grammar and later, it is translated into a SQL query by the use of NLP (Natural Language Processing). The interface makes use of RNN (Recurrent Neural Networks) to convert the natural language input given by the user to the Structured Query language. The network is trained on large number of datasets from the WikiSQL, which makes sure that the accuracy of the model is good enough for real time use in the industry and also for the personal use. Stanford NLP and self-made datasets are also used to an extent to enhance the efficiency and optimise the query formation process. Finally, a database management system is used to find the result set with its own specialized optimization and planning techniques. The database is implemented in the JSON format to reduce the query processing time when the database is too large. The database is also organised in a tiered architecture format so that the query execution time is reduced.