Text to SQL Query Conversion Using Deep Learning: A Comparative Analysis

Abstract

In relational databases, a significant amount of world’s knowledge is stored. The ability of users to retrieve facts from a database is limited due to a lack of understanding of query languages. Intersection of Natural Language Processing (NLP) and human-computer intersection is the Natural Language Interface (NLI) which provides humans to interact with computer through the use of natural language. Here, NLI applied to relational databases translating Natural Language Queries to Structured Query Language (SQL). The proposed system is a NLP based system for converting natural language queries to database queries using Deep Neural Network. Spider, a new large-scale, cross-domain semantic parsing and text-to-SQL data set annotated by 11 college students released on November 2018 is used here. The system uses a deep learning framework such as basic sequence to sequence (Encoder-Decoder Model) and sequence to sequence plus attention. The model takes word embeddings of the query as the input and then sequence to sequence modeling is applied. The system is evaluated on the two models based on the exact matching and hardness criteria of questions. The evaluation results on Spider dataset shows that the proposed two deep learning model improves the results of text-to-SQL task. Spider dataset poses a major challenge for future research.

Keywords: Deep Learning; Natural Language Interface (NLI); Natural Language Processing (NLP); Semantic Parsing; Seq2Seq; Seq2Seq+Attention