BriefGPT.xyz
Sep, 2021
用位置掩蔽改进文档布局感知理解
Position Masking for Improved Layout-Aware Document Understanding
HTML
PDF
Anik Saha, Catherine Finegan-Dollak, Ashish Verma
TL;DR
本文探讨了一种新的预训练任务,即使用位置嵌入来提高LayoutLM的性能,与使用仅语言掩码的模型相比,使用位置掩码的模型在表单理解任务上的表现提高了超过5%。
Abstract
natural language processing
for document scans and
pdfs
has the potential to enormously improve the efficiency of business processes. Layout-aware word embeddings such as
→