| The emergence of building information modeling(BIM)technology has promoted the development of Compliance checking to a certain extent.At present,the research on Compliance checking based on the BIM model is mostly a semi-automatic method or a combination of manual and automatic methods.The introduction of artificial intelligence technology into Compliance checking is expected to realize the full automatic review of the BIM model.Generally,the Compliance checking based on the BIM model is divided into four steps: specification translation,BIM model preparation,rule execution,and report generation.Among them,the automatic classification of specification terms into predefined categories is the prerequisite for the realization of specification translation,that is,the text classification of design specifications.Therefore,this study introduces the natural language processing technology in the field of artificial intelligence into the field of structural design to assist the structural design review based on the BIM model.The main work is as follows:Firstly,this paper summarizes the research status of the Compliance checking,text representation,and text classification at home and abroad,and points out the few existing problems of normative classification in a Compliance checking.Aiming at the research problems,this paper introduces the research content,technical route,and architecture arrangement of this paper,and expounds in detail on the relevant technical theories involved in the text classification of structural design specifications based on natural language processing,as well as how to establish the corpus of structural specifications in the field of construction design.Then,the text classification method of structural design specification is designed in the following aspects:(1)In the aspect of machine learning,this paper proposes a research method of structural design specification text classification based on a machine learning algorithm.The research process includes(1)data collection;(2)Text preprocessing;(3)Word embedding model;(4)Classifier selection;(5)Evaluation and results of text classification.Three spatial models(TFIDF,Word2 Vec,and Doc2Vec)and six machine learning algorithms(Naive Bayes,K-nearest neighbor,Support vector machine,Logistic regression,Gradient boosting decision tree,Random forest)were implemented and evaluated under different thresholds.(2)In the aspect of deep learning,given the problems that the global information cannot be considered in the text representation process and the machine learning algorithm has poor learning ability and simple model structure,this paper proposes a research method of structural design specification text classification based on a deep learning algorithm.Data vectorization is carried out through one hot and glove word vector training model considering global information.After that,three basic deep learning models(LSTM,CNN,and GRU)that can make more intelligent in-depth learning and decision-making on data and six improved deep learning models(Bi LSTM,Bi GRU,CNN-LSTM,CNN-GRU,CNN-Bi LSTM,CNN-Bi GRU)will be used for feature extraction and model training,And import Softmax classifier for classification.(3)In the aspect of deep learning + transfer learning,because the five text representation models previously studied can not solve the problems of polysemy,this paper proposes a research method of structural design specification text classification based on the ALBERT pretraining model.ALBERT pre-training model is a transfer learning model which is more suitable for small data sets based on the BERT transfer learning model which can solve the polysemy of one word.Through the transfer learning model and the two-way gated loop unit(Bi GRU)combined with the attention mechanism,the normative classification model——ALBERT+Bi GRU is used in this paper.The research results show that the ALBERT +Bi GRU model proposed in this paper has higher accuracy than the above model based on machine learning and deep learning in the research of structural design specification text classification,with an accuracy rate of 82.6%,a recall rate of 78.7% and an F1 value of 80.6%.Finally,combined with the above basic model and improved model algorithm,this paper studies the text classification of the structural design specification,designs the GUI Graphical user interface(hereinafter referred to as "graphical user interface design")and realizes the page visualization of the text classification process in this field.This paper introduces natural language processing technology into the text classification research of structural design specifications in the architectural field.To a certain extent,this research realizes the automatic classification of entities according to predefined categories in BIM compliance checking,provides a classification method for future research in this field,and promotes the development of BIM compliance automatic checking. |
