TableBench
收藏资源简介:
TableBench是由北京航空航天大学等机构创建的综合性表格问答数据集,包含886个样本,覆盖18个领域。该数据集通过系统分析真实世界表格应用的挑战,定义任务复杂性,并结合人工和自动化方法构建。TableBench主要用于评估和提升大型语言模型在事实检查、数值推理、数据分析和可视化等任务上的性能,旨在弥合学术基准与工业应用之间的差距。
TableBench is a comprehensive tabular question answering dataset developed by Beihang University and other institutions. It contains 886 samples spanning 18 distinct domains. This dataset is constructed by systematically analyzing the challenges of real-world tabular applications, defining task complexity, and integrating both manual and automated approaches. TableBench is primarily designed to evaluate and improve the performance of large language models (LLMs) on tasks including fact-checking, numerical reasoning, data analysis and visualization, with the goal of bridging the gap between academic benchmarks and industrial applications.

- 1TableBench: A Comprehensive and Complex Benchmark for Table Question Answering北京航空航天大学 · 2024年



