TableBank
收藏资源简介:
为了解决对标准开放域表基准数据集的需求,作者提出了一种新颖的弱监督方法来自动创建 TableBank,该方法比现有的用于表分析的人工标记数据集大几个数量级。与传统的弱监督训练集不同,我们的方法不仅可以获得大规模的训练数据,而且可以获得高质量的训练数据。如今,网络上有大量的电子文档,例如 Microsoft Word (.docx) 和 Latex (.tex) 文件。这些在线文档本质上在其源代码中包含表格的标记标签。直观地说,可以通过在每个文档中使用标记语言添加边界框来操作这些源代码。对于 Word 文档,可以在标识每个表格的边界的地方修改内部 Office XML 代码。对于 Latex 文档,也可以在识别表格边界框的地方修改 tex 代码。通过这种方式,可以为商业文档、官方填写、研究论文等各种领域创建高质量的标记数据,这对于大规模的表格分析任务非常有利。 TableBank 数据集总共包含 417,234 个高质量标记表及其在各个领域的原始文档。
To address the demand for standard open-domain table benchmark datasets, the authors proposed a novel weakly-supervised method to automatically create TableBank, which is several orders of magnitude larger than existing manually annotated datasets for table analysis. Unlike traditional weakly-supervised training datasets, our method can acquire not only large-scale training data but also high-quality training data. Nowadays, there are vast amounts of electronic documents on the Internet, such as Microsoft Word (.docx) and LaTeX (.tex) files. Essentially, these online documents contain markup tags for tables within their source code. Intuitively, these source codes can be manipulated by adding bounding boxes using markup languages within each document. For Word documents, the internal Office XML code can be modified at the positions where the boundaries of each table are identified. For LaTeX documents, the .tex source code can also be modified at the positions where the bounding boxes of tables are recognized. In this way, high-quality annotated data can be created for various domains such as business documents, official forms, research papers and more, which is highly beneficial for large-scale table analysis tasks. The TableBank dataset contains a total of 417,234 high-quality annotated tables along with their original source documents across various domains.

- TableBank数据集首次发表,由腾讯和北京大学联合发布,旨在提供一个大规模的表格检测和识别数据集。
- TableBank数据集首次应用于学术研究,特别是在文档分析和信息提取领域,推动了相关技术的进步。
- TableBank数据集被广泛应用于多个国际竞赛和挑战赛,进一步验证了其在表格处理任务中的有效性。
- 1TableBank: Table Benchmark for Image-based Table Detection and RecognitionUniversity of Waterloo, Microsoft Research Asia · 2019年
- 2TableNet: An Approach for Determination of Fine Similarities in Tables of a Similar TypeIndian Institute of Technology Guwahati · 2020年
- 3Deep Learning for Table Detection and Structure Recognition: A SurveyUniversity of Science and Technology of China · 2021年
- 4Table Structure Recognition with Conditional AttentionUniversity of Science and Technology of China · 2020年
- 5TabStruct-Net: Structure-based Action Recognition in Tabular DataUniversity of Science and Technology of China · 2021年



