我国城市轨道交通标准指标自动提取技术研究

China's Urban Rail Transit Standard Indicators Automatic Extraction Technology Research

我国城市轨道交通标准化工作发展迅速,已形成大量标准,目前依赖专业技术人员开展分析对标,整体效率低下。本文提出基于深度学习的我国城轨标准指标提取技术,实现标准文本中各类指标的自动提取,能够高效、准确地为下游任务提供数据支持,是标准数字化技术中重要的基础技术之一,更是城轨标准化工作的创新探索。对于城轨标准中常见的两类指标,文本描述类指标及表格类指标,分别构建识别及提取模型。特别是针对复杂表格中的指标抽取,提出一种基于图注意力网络的表格类指标信息抽取方法。经验证,本文提出的技术能够自动识别录入系统的城轨标准文本中的指标并提取出来。

With the rapid development of China's urban rail transit standardization work,a large number of standards have been formed,which currently rely on professional and technical personnel to carry out analysis,resulting in low efficiency.A China's urban rail transit standard indicators automatic extraction technology based on deep learning is proposed to achieve automatic extraction of standard indicators and provide data support for downstream tasks efficiently and accurately,which is one of the important basic technologies in standard digital technology,and also an innovative exploration of urban rail transit standardization.Identification and extraction models are constructed respectively for the text-based indicator and the tablebased indicator which are the two common indicators in urban rail transit standards.A table-based indicators extraction method based on graph attention network is proposed for the extraction of indicators in complex tables.It is verified that the technique proposed in this paper can automatically identify the indicators in urban rail transit standards entered into the system.