施工方案类别及文本分类模型实现研究分类

Construction scheme category and text classification model realization research classification

摘要

施工方案是施工组织设计的核心环节。运用自然语言处理技术，从非结构化文本中提取施工方案，并予以审阅和审查，可有效提高施工方案的审查效率，提高编制质量，同时有利于发现施工方案中潜在的安全和质量风险，以便施工过程加以预警。从非结构化文本提取施工方案，需要明确不同类型的施工方案的内容构成，对相关段落进行内容归类。针对非结构化施工方案段落内容归类问题，在深入研究工程施工方案类别及内容构成框架的基础上，以城市管网工程施工组织设计段落为样本，进行了施工组织设计段落内容分类，提出了融合Albert、TextRCNN的段落文本分类模型，该模型采用Albert预训练语言模型进行词嵌入，将生成的词向量输入到TextRCNN分类器中完成文本分类，准确率提高0.79%,试验表明：结合Albert的TextRCNN可以有效对施工组织设计段落进行内容分类，为进一步施工方案提取提供基础。

Abstract

Construction scheme is one of the most important contents of construction organization design. The application of natural language processing to extract and review construction schemes from unstructured texts can improve review efficiency, enhance preparation quality, and identify potential safety and quality risks for early construction warnings. To extract construction schemes from unstructured texts, it is necessary to first clarify the content composition of different types of construction schemes and classify the contents of relevant paragraphs. Aiming at the classification of unstructured construction scheme paragraphs, classification of construction organization design paragraphs of urban pipe network engineering was taken as samples on the basis of in-depth study of construction scheme categories and content composition framework, and proposes a paragraph text classification model integrating Albert and TextRCNN. This model uses Albert pretraining language model for word embedding, and input the generated word vector into TextRCNN classifier to complete text classification. It outperforms Albert-TextCNN, which has the best classification effect among the other three models, and the accuracy increases by 0.79%. The experiment shows that: TextRCNN combined with Albert can effectively classify the contents of the construction organization design paragraphs, providing a basis for further construction scheme extraction.

关键词

施工方案 / 段落文本 / Albert / TextRCNN

Key words

construction scheme / paragraph text / Albert / TextRCNN

Author summay

杜润隆(1990—), 男, 高级工程师, 学士, 主要从事数字化研究。E-mail: [e-mail]

基金资助

中国长江三峡集团有限公司科研项目(202103551)