Research On Construction Safety Accidents Based On Text Mining And CBR

The construction industry is a pillar industry of the national economy,but it is also a field with frequent and high occurrence of safety problems,and even similar accidents are staged one after another.It has been an academic focus of engineering safety governance to investigate and study the occurrence pattern of construction safety accidents,and to mine accident information from similar cases,for which this study adopts a combination of text mining and Word2Vec model to build a construction safety accident-specific domain This study uses a combination of text mining and the Word2Vec model to build a dictionary of construction safety accidents,to extract causal factors and parametrically represent the structured storage and risk identification of cases,and to provide pre-warning and post-response measures for target cases through case inference.There is currently a lack of specialised dictionaries in this field in China.In this paper,we use the investigation reports of construction safety accidents as the corpus,firstly,we pre-process the corpus,secondly,we use the TF-IDF algorithm to screen out the seed word set,and then we use the Word2Vec model to train the word vector,then we use the similarity calculation to determine the domain candidate words,and we complete the construction of the domain lexicon through screening,and we use SVM and the plain Bayesian model to test the lexicon word separation.The lexicon was tested by SVM and plain Bayesian model,and the lexicon was tested to be very effective.On this basis,the accident investigation report is analysed by text mining to extract the key items,and the cause of the accident is extracted by retracing the performance of the feature items in the accident report.On the basis of the above,the TM-CBR model is proposed for the study of construction safety accidents.The attribute indicators for this paper were first identified through the literature review method,followed by the analysis of the attribute indicators by classifying them into numerical,character and text types,and using the weights associated with the local similarity of each feature,the global similarity between cases was calculated using the AHP method,and case correction and case retention were carried out based on the results.Finally,this study carried out an empirical analysis to collect and sort out the cases of building construction safety accidents between 2015 and 2021,and collated 1094 cases of building construction safety accidents according to the extracted attribute indicators to form the case base for this study.Based on the TM-CBR inference model,the case search and matching was carried out to retrieve and warn construction safety incidents beforehand,control them during the incident and deal with them precisely afterwards.This research can effectively analyse the laws of construction safety accidents and provide theoretical reference and practical guidance for safety management.