SQLSketch-TVC: Type, value and compatibility based approach for SQL queries

Karam Ahkouk,Mustapha Machkour
DOI: https://doi.org/10.1007/s10489-022-03587-0
IF: 5.3
2022-06-05
Applied Intelligence
Abstract:Understanding the complexity of the translation of Natural Language (NL) sentences to SQL queries becomes an essential part in the resolution process. The majority of the proposed models either focus on simple queries or suffer when exposed to unseen domains or new schemas structures; This can be understood as the greater part of solutions are based on limited datasets or treat the problem in an end-to-end perspective. Our previously proposed model which is SQLSketch that provides an intelligent method for handling complex queries was able to outperform all the state-of-the-art models on the GreatSQL dataset. This paper addresses the problem of translating NL sentences to SQL queries in an effective way by leveraging our previous SQLSketch model with a type aware layer, a values classification method as well as a compatibility based module that enhance the quality of the predicted items (SQLSketch-TVC). We evaluate the new model using the Components and Exact matching metrics. The results show that SQLSketch-TVC outperforms the other models on all SQL components and provides a novel way for inferring values from the input Question.
computer science, artificial intelligence
What problem does this paper attempt to address?