Enhancement of Natural Language to SQL Query Conversion using Machine Learning Techniques

In the age of information explosion, there is a huge data that is stored in the form of database and accessed using various querying languages. The major challenges faced by a user accessing this data is to learn the querying language and understand the various syntax associated with it. Query given in the form of Natural Language helps any naive user to access database without learning the query languages. The current process of conversion of Natural Language to SQL Query using a rule-based algorithm is riddled with challenges - identification of partial or implied data values and identification of descriptive values being the predominant ones. This paper discusses the use of a synchronous combination of a hybrid Machine Learning model, Elastic Search and WordNet to overcome the abovementioned challenges. An embedding layer followed by a Long Short-Term Memory model is used to identify partial or implied data values, while Elastic Search has been used to identify descriptive data values (values which have lengthy data values and may contain descriptions). This architecture enables conversion systems to achieve robustness and high accuracies, by extracting meta data from the natural language query. The system gives an accuracy of 91.7% when tested on the IMDb database and 94.0% accuracy when tested on Company Sales database.