Scholarlink

Question Answering on Tabular Data has garnered significant interest within the Natural Language Processing community due to its real-world applicability in extracting insights from structured data, thereby enhancing data accessibility and decision-making. This paper introduces an approach leveraging large language models and prompt engineering to translate natural language queries into executable code (Python and SQL). We explore and combine various strategies to optimize performance. Experiments conducted on a Vietnamese translation of the test data from SemEval 2025 Task 8: DataBench, Question-Answering over Tabular Data demonstrate that framing the task as a text-to-Python code generation problem, utilizing the capabilities of GPT-4o, yields superior performance across two sub-tasks.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Vietnamese Question Answering on Tabular Data via Large Language Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nguyen Minh Son,
  • Bui Hong Son,
  • Le Minh Hung,
  • Vo Tuan Kiet,
  • Dang Van Thin

摘要

Question Answering on Tabular Data has garnered significant interest within the Natural Language Processing community due to its real-world applicability in extracting insights from structured data, thereby enhancing data accessibility and decision-making. This paper introduces an approach leveraging large language models and prompt engineering to translate natural language queries into executable code (Python and SQL). We explore and combine various strategies to optimize performance. Experiments conducted on a Vietnamese translation of the test data from SemEval 2025 Task 8: DataBench, Question-Answering over Tabular Data demonstrate that framing the task as a text-to-Python code generation problem, utilizing the capabilities of GPT-4o, yields superior performance across two sub-tasks.