An Intelligent Natural Language Interface for Querying Databases based on Deep Learning.

Natural Language Interfaces to Databases (NLIDBs) aim to simplify access to structured data by allowing users to express database queries in natural language. Despite significant progress in this field, many existing approaches still struggle with handling complex queries, adapting to diverse database schemas, and generalizing across domains. Recent advances in deep learning offer promising solutions by learning effective mappings between natural language expressions and structured query languages. In this work, we first review and analyze recent encoder-decoder-based approaches, highlighting their strengths and limitations. We then introduce Chat-SQL, an intelligent deep learning-based NLIDB framework designed to translate natural language questions into executable SQL queries. The proposed system employs a modular architecture that integrates natural language preprocessing, semantic representation, schema-aware encoding, and structured SQL query generation. By framing the task as a Text-to-SQL translation problem, Chat-SQL enhances robustness to linguistic variability and reduces user's reliance on technical knowledge of database schemas. Our approach leverages a hybrid architecture that combines sketch-based query representations with large language models (LLMs). The system architecture is detailed, including its sketch selection module, LLM-based slot-filling mechanism, and dialogue state tracking component, which together enable accurate and context-aware query construction. Experimental evaluations on standard benchmark datasets, complemented by user-centered studies, demonstrate that Chat-SQL outperforms baseline methods and conventional single-turn systems in terms of query correctness, multi-turn contextual understanding, and overall user satisfaction. Its modular design and strong generalization capabilities make it a flexible and effective solution for natural language database querying. [ABSTRACT FROM AUTHOR]