Bridging the Gap: Enabling Natural Language Queries for NoSQL Databases through Text-to-NoSQL Translation

The paper introduces TEND (Text-to-NoSQL Dataset), a benchmark for translating natural language queries into executable MongoDB queries using aggregation pipelines, specifically designed for schema-less document stores. TEND comprises 1,210 tasks that evaluate the ability to reason about non-relational data structures, with a focus on unique challenges such as nested arrays and dynamic keys, contrasting with traditional SQL-to-MQL models. Additionally, the authors present SAG (Schema-as-Data Grounding), a solver that enhances query generation by grounding paths and values from stored documents, highlighting the distinct difficulties LLMs face in handling Text-to-NoSQL compared to NL2SQL tasks.