施工组织设计文档智慧辅助审查中的文本分类问题研究

Document Classification in Intelligent Aided Review of Construction Organization Design Documents

施工组织设计是指导工程建设全过程活动的技术、经济和组织的综合性文件,随着自然语言处理(natural language processing,NLP)等人工智能技术的发展,针对施工组织设计文档智慧辅助审查中基础性工作:文本分类问题开展研究。为实现施工组织设计文本的自动分类,运用Word2vec词嵌入技术对文本进行向量化表示,基于双向长短时记忆网络(bi-directional long short-term memory,Bi-LSTM)捕捉文本上下文序列信息,融入Attention机制,提取文本有效信息,采用softmax激活函数分类。结果表明:Attention Bi-LSTM在房建数据集上达到0.97的准确率、召回率以及F值,整体分类效果在正确率、宏平均、加权平均上均优于其他模型。融入Attention机制的Bi-LSTM文本分类模型通过双向捕获文本的特征并利用Attention机制提取有效信息,达到了联合优化的作用,提高了模型的分类性能。

Construction organization design is a comprehensive technical,economic and organizational document that is used to guide the whole process of project construction.With the development of artificial intelligence technologies such as natural language processing(NLP),basic work in intelligent auxiliary review of construction organization design documents:text classification is studied.In order to realize the automatic classification of municipal construction organization design text,Word2vec word embedding technology was used to vectorize text.In order to extract text information effectively,bi-directional long short-term memory(Bi-LSTM)was used to capture text context sequence information,integrating Attention mechanism,and softmax activation function was used to classify.The results show that Attention Bi-LSTM achieves the accuracy rate,recall rate and F value of 0.97 on the housing construction dataset,and the overall classification effect is better than other models in terms of accuracy rate,macro average,and weighted average.The Bi-LSTM text classification model integrated with the Attention mechanism achieves the effect of joint optimization by capturing the features of the text in both directions and extracting effective information using the Attention mechanism,which improves the classification performance of the model.