iTBLS

iTBLS数据集由佐治亚理工学院AI虚拟助理实验室创建，包含4000个来自科学文章的表格，旨在通过AI驱动的多任务表格能力促进人机协作问题解决。该数据集扩展了交互范围，包括数学推理、自然语言操作和基于自然语言对话的现有表格扩展。数据集通过分解交互为解释、修改和生成三个任务，支持更复杂的对话交互框架，如比较、确定绝对和相对位置及数学推理。此外，iTBLS利用arXiv作为主要信息源，引入了现有数据集中未包含的新颖且丰富的信息源。

The iTBLS dataset was created by the AI Virtual Assistant Lab at the Georgia Institute of Technology. It contains 4,000 tables extracted from scientific articles, and is designed to facilitate human-AI collaborative problem-solving via AI-powered multi-task table capabilities. This dataset expands the scope of interactive scenarios, covering mathematical reasoning, natural language manipulation, and existing table expansion tasks grounded in natural language dialogue. By decomposing table interactions into three core tasks: explanation, modification, and generation, the dataset supports more sophisticated conversational interaction frameworks such as comparison, determining absolute and relative positions, and mathematical reasoning. Additionally, iTBLS takes arXiv as its primary information source, introducing novel and rich information resources not covered by existing datasets.