Inventory Management: Dataset for Scalable Text-to-SQL Systems

With the creation of massive datasets like Spider, natural language interfaces for databases have advanced significantly. However, important fields like inventory management are still under-represented. These topics are critical because non-technical users utilize inventory databases globally and would greatly benefit from the ability to retrieve information using natural language queries. This work proposes a specific Text-to-SQL dataset designed for inventory management. Our dataset comprises 50,000 natural language query-SQL pairings across inventory databases with over a million records. Using this dataset, we assess state-of-the-art models (e.g., GPT-4), uncovering significant performance issues and highlighting the need for new domain-specific models. This study contributes a valuable resource to the development of Text-to-SQL programs in inventory management.