一种基于大模型和检索增强的表格问答方法
By fine-tuning the CodeLLaMA-7B-Instruct-hf model using LoRA and enhancing the multimodal knowledge base, the accuracy issues of table-based question answering and multi-table queries in Chinese software development scenarios were resolved, achieving efficient and accurate data analysis and lowering the query threshold.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV
- Filing Date
- 2025-07-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing table-based question-answering technologies suffer from low accuracy, difficulty in adapting to multiple table structures, and lack of domain knowledge in Chinese software development scenarios, making it difficult to meet the needs for efficient and accurate data analysis.
We constructed a Text-to-SQL model, LLMforTable, using the CodeLLaMA-7B-Instruct-hf large model combined with LoRA parameter fine-tuning. We then generated a structured Prompt to improve accuracy by using a multimodal knowledge base and an incremental difficulty training strategy, combined with Sentence-BERT and IVF-PQ algorithms.
In Chinese software development scenarios, it significantly improves the accuracy of SQL generation, reduces training resource consumption, shortens training time, and enables non-technical users to directly obtain database information through natural language, thus lowering the query threshold.
Smart Images
Figure CN120929485B_ABST