AnnotatedTables
收藏资源简介:
AnnotatedTables是由爱荷华州立大学计算机科学系创建的大型表格数据集,包含32,119个数据库,总计405,616个有效的SQL程序。该数据集利用大型语言模型(LLMs)自动生成注释,解决了传统人工注释的扩展瓶颈。数据集的构建始于跨领域的实际数据科学应用中的多样化表格数据,通过精心设计的提示,指导LLMs使用零样本学习合成SQL代码。AnnotatedTables不仅支持查询执行,还适用于各种研究目标,如SQL到Rel程序的翻译和表格分类模型的评估。此数据集展示了LLMs在自动化大量多样化表格数据注释方面的潜力,适用于解决数据库管理、查询优化等领域的实际问题。
AnnotatedTables is a large-scale tabular dataset created by the Department of Computer Science at Iowa State University, which encompasses 32,119 databases and a total of 405,616 valid SQL programs. This dataset utilizes Large Language Models (LLMs) to automatically generate annotations, addressing the scalability bottleneck inherent in traditional manual annotation workflows. The construction of AnnotatedTables begins with diverse tabular data sourced from real-world data science applications across multiple domains. Through carefully crafted prompts, LLMs are guided to synthesize SQL code via zero-shot learning. AnnotatedTables not only supports query execution but also caters to a wide range of research goals, including SQL-to-Rel program translation and the evaluation of tabular classification models. This dataset showcases the potential of LLMs in automating annotations for large volumes of diverse tabular data, and is applicable to solving practical problems in fields such as database management and query optimization.

- 1AnnotatedTables: A Large Tabular Dataset with Language Model Annotations爱荷华州立大学计算机科学系 · 2024年



