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

TL;DR

This work introduces the Text-to-NoSQL task to translate natural language queries into NoSQL queries, addressing a key usability gap for NoSQL systems. It presents TEND, a large-scale, open-source benchmark, and SMART, a four-stage SLM+RAG framework that predicts schemas, generates and refines NoSQL queries, and optimizes them using execution feedback. The authors propose a semi-automatic data-construction pipeline, a suite of novel evaluation metrics, and extensive experiments showing SMART significantly outperforms strong baselines while ablations confirm each component’s value. The work lays a foundation for accessible NoSQL querying, offering a scalable pathway to real-world NL interfaces for diverse NoSQL platforms and datasets.

Abstract

NoSQL databases have become increasingly popular due to their outstanding performance in handling large-scale, unstructured, and semi-structured data, highlighting the need for user-friendly interfaces to bridge the gap between non-technical users and complex database queries. In this paper, we introduce the Text-to-NoSQL task, which aims to convert natural language queries into NoSQL queries, thereby lowering the technical barrier for non-expert users. To promote research in this area, we developed a novel automated dataset construction process and released a large-scale and open-source dataset for this task, named TEND (short for Text-to-NoSQL Dataset). Additionally, we designed a SLM (Small Language Model)-assisted and RAG (Retrieval-augmented Generation)-assisted multi-step framework called SMART, which is specifically designed for Text-to-NoSQL conversion. To ensure comprehensive evaluation of the models, we also introduced a detailed set of metrics that assess the model's performance from both the query itself and its execution results. Our experimental results demonstrate the effectiveness of our approach and establish a benchmark for future research in this emerging field. We believe that our contributions will pave the way for more accessible and intuitive interactions with NoSQL databases.

Figures (4)

• Figure 1: An example of Text-to-NoSQL involves a Text-to-NoSQL model converting a user's natural language query into a NoSQL query, which is then executed in the corresponding NoSQL database to obtain the desired result.
• Figure 2: Construction pipeline for the TEND dataset. Through an automated process, the conversion from a Text-to-SQL dataset to a Text-to-NoSQL dataset can be achieved. Specifically, (i) the conversion from SQL databases to NoSQL (e.g., MongoDB) databases is accomplished through algorithmically programmed software; (ii) by inputting examples into the most advanced LLM (such as GPT-4) to obtain examples of Chain-of-Thought, namely Advanced COT, which assists a second-tier LLM (such as GPT-3.5) in reasoning, thus achieving automated generation of NoSQL queries, feedback generation, and query debug; (iii) the expansion of dataset questions is implemented using Multi-LLM.
• Figure 3: The working pipeline of our proposed SMART framework, where (i) SLM-based Schema Prediction predicts the required NoSQL schemas by fine-tuning the SLM. (ii) SLM-based Query Generation generates initial NoSQL queries by fine-tuning the SLM. (iii) Predicted Schema-driven and Retrieved Example-driven Query Refinement is responsible for refining the NoSQL queries generated by the SLM. The RAG technique used here is detailed in Section \ref{['sec:query_refine']}. (iv) Execution Result-based Query Optimization optimizes the refined NoSQL queries based on their execution results.
• Figure 4: Parameter study. These are the variation curves of SMART under different numbers of retrieval examples and different metrics. The vertical axis represents the model's accuracy under a specific metric, and the horizontal axis represents the number of retrieval examples.