Reproduce “ChatBI: Towards Natural Language to Complex Business Intelligence SQL” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jinqing Lian, Xinyi Liu, Yingxia Shao, Yang Dong, Ming Wang, Zhang Wei, Tianqi Wan, Ming Dong and 1 more

ChatBI introduces a system for converting natural language queries into complex SQL within Business Intelligence environments, specifically designed to handle multi-round dialogues, large ambiguous schemas, and intricate business logic. The system demonstrates superior useful execution accuracy and token efficiency compared to existing LLM-based NL2SQL methods on real-world BI datasets.

Key insights

  • Decomposing the Natural Language to SQL problem by having the LLM generate an intermediate JSON representation simplifies the LLM's task and significantly improves accuracy for complex Business Intelligence (BI) queries.
  • Strategically utilizing smaller, specialized models for specific sub-problems like multi-round dialogue context management and single view selection enhances efficiency and reduces token costs in production environments.
  • Integrating established database concepts such as 'views' effectively manages schema complexity and ambiguity, alleviating token limits and improving schema linking accuracy for wide BI tables.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.