FI-NL2PY2SQL: Financial Industry NL2SQL Innovation Model Based on Python and Large Language Model

Abstract

With the rapid development of prominent models, NL2SQL has made many breakthroughs, but customers still hope that the accuracy of NL2SQL can be continuously improved through optimization. The method based on large models has brought revolutionary changes to NL2SQL. This paper innovatively proposes a new NL2SQL method based on a large language model (LLM), which could be adapted to an edge-cloud computing platform. First, natural language is converted into Python language, and then SQL is generated through Python. At the same time, considering the traceability characteristics of financial industry regulatory requirements, this paper uses the open-source big model DeepSeek. After testing on the BIRD dataset, compared with most NL2SQL models based on large language models, EX is at least 2.73% higher than the original method, F1 is at least 3.72 higher than the original method, and VES is 6.34% higher than the original method. Through this innovative algorithm, the accuracy of NL2SQL in the financial industry is greatly improved, which can provide business personnel with a robust database access mode.

Keywords

• LLM
• NL2SQL
• pre-training
• prompt
• Python